This article provides recommendations for the best ways to collect and preserve evidence using eyewitness identification procedures. The recommendations are based on decades of research by dozens of memory and social science researchers. Many of the recommendations are included in the recent report on the Robert Sophonow case in Manitoba and in a 1999 U.S. National Institute of Justice document distributed to all police services in the U.S. The recommendations are part of training programs for several police services in Canada, the U.S., and around the world, and experienced criminal investigators will recognize many of them as ideas they have used in their own cases. Our goal is to put all of these ideas in one place so that more investigators can take advantage of them and achieve a greater goal of maximizing the rate of accurate eyewitness identifications while minimizing the rate of inaccurate choices.

Criminal investigators know that it often takes many pieces of converging evidence to solve a complex case. Freshly-schooled recruits and veteran investigators alike are trained to search for, detect, collect, and preserve "obvious" physical evidence such as weapons and stolen property, as well as trace physical evidence such as fibres, hairs, fingerprints, blood, and semen. Few police officers, lawyers, scientists, or people in general would question the importance of using the best procedures available to obtain and preserve such evidence, not to mention adherence to relevant statutory and case law. But eyewitness evidence ­ the testimony of victims, witnesses, and suspects, and perhaps the identification of a suspect from a live or photo lineup ­ is a common form of evidence that presents some unique problems to investigators, which in turn can lead to especially serious errors in court.

An analysis of actual cases in the U.S. has revealed that the mistaken identification of the wrong person by victims and witnesses to a crime is the single most common error leading to the arrest and conviction of innocent people there (U.S. National Institute of Justice, 1996), and several cases in Canada suggest similar problems here (e.g., R. v. McGuiness, Ballantyne & Ballantyne, 1997; R. v. Sophonow, 1985). On the other side of the coin, potentially valuable eyewitness information is sometimes overlooked by not employing optimal procedures. In both cases, the actual perpetrator of the crime is still at large, either because the wrong person has been identified or because no evidence has been produced.

This article describes a set of recommendations designed to minimize the rate of eyewitness errors, while at the same time maximizing the rate of accurate identifications. The recommendations are supported by over 20 years of research, are consistent with the U.S. Guide for Law Enforcement on Eyewitness Evidence (U.S. National Institute of Justice, 1999), are the basis of training curricula for several police services in Canada and the U.S. (e.g., Ottawa-Carleton, York Region, the entire state of New Jersey), and are in use by thousands of police officers in Canada and around the world (see Kebbell, 2000 for a favorable comparison between practices in the United Kingdom and these recommendations). Our goal is to present these ideas and their rationale so that even more investigators can take advantage of them (additional related articles are listed in Note 1).

A Psychological Assessment of the Eyewitness Evidence Predicament
It is not a ground breaking statement to say that memory is not perfect. Anyone who has forgotten a name, misplaced the car keys, or scored less than 100% on an exam knows that not everything people initially perceive can be recalled accurately at a later date. But this apparently simple notion becomes very complicated in the case of an eyewitness's memory for a criminal event or traffic accident. The resolution of a case can depend heavily on such information and the consequences of an error can be serious. It is not surprising, therefore, that there exists a large body of research findings in this interesting overlap between basic memory research in psychology and very applied problems in criminal investigations and criminal law. Here we briefly describe some of the factors that can affect memory at the three stages of human information-processing (acquisition, storage, and retrieval) involved in eyewitness cases, although the remainder of this article will focus on a different approach to the problem.

First, decades of research reveals that there can be problems with people's ability to accurately perceive and thereby "acquire" all of the details of an event right from the start. A person's state of heightened physiological arousal in a threatening situation, at one extreme, or their failure to even notice that a crime was occurring at the other extreme (e.g., a teller cashing a forged check), are examples of potential perception problems. Other examples could include the viewing conditions at the time of the event, the eyesight of the witness, and possibly the race of the witness compared to that of the offender.

Second, assuming that some of the details from the initial event have been perceived correctly, those details face competition in the person's memory from other information to which that person is exposed after the event. The potential for such "memory interference" from several sources (e.g., other witnesses and media reports) has been well documented. In many cases, successful strategies to reduce the problem have been developed (e.g., separating witnesses before they can speak with one another, requesting that witnesses avoid speaking with the media or imposing an outright ban on media releases, and minimizing the time between the event and the retrieval attempt).

Third, even if the accurately perceived information remains intact in memory until its retrieval is requested in a prompt investigative interview or a lineup viewing, inappropriate retrieval strategies can lead to an incomplete statement or an inaccurate description of the offender. Sometimes insufficient effort is requested from and/or exerted by a witness to exhaustively search their memory for all available information, or a description-eliciting technique asks for a level of detail that is inconsistent with the overall impression the witness might have of the offender's appearance.
So, memory is prone to error at several stages of information processing. Memory factors must therefore be taken into account when assessing the potential accuracy of witnesses' statements or their ability to accurately identify a perpetrator. But the advantage to law enforcement of recognizing these potential memory shortcomings is limited, which leads to a useful distinction between two categories of eyewitness evidence factors.

System Variables vs. Estimator Variables
Not all of the potentially negative influences on the quantity and quality of eyewitness evidence are as vulnerable to human shortcomings as others, a fact highlighted by the distinction between "system variables" and" estimator variables" first discussed over 20 years ago by one of the current authors (Wells, 1978). Estimator variables are factors over which the police and the justice system have relatively little control, so that their actual values and possible effects can only be estimated as opposed to controlled. Some common examples of estimator variables have been mentioned above (e.g., the amount of attention paid by the witness at the time of the event, the quality of the viewing conditions, the length of time between the event and the taking of a statement). In addition to these primarily memory-based problems, there are other estimator variables such as witnesses' expectations of what they're supposed to do, inferences they might make about the identity of the offender, and possibly their overzealousness to cooperate on the one hand or to intentionally stonewall an investigation on the other.

In contrast, much of the research in the eyewitness area, especially recently, has focused on system variables ­ those factors over which the police and the justice system have at least some degree of control. Some common examples have been mentioned above (e.g., whether or not witnesses are separated before they have the opportunity to exchange information with one another and the techniques used by an investigator to elicit a description from a witness). Other system variables will be discussed at length below (e.g., the construction and presentation of identification lineups). The goal of this article is to provide practical recommendations for police to maximize the effectiveness of eyewitness evidence, so the focus here is on system variables.

A Trace-Evidence Perspective
The term "trace evidence" is usually reserved for physical evidence such as fingerprints, blood, or fibres. It is useful, however, to also think of eyewitness memory as a form of trace evidence, an idea also introduced by one of the current authors (Wells, 1995). Like physical trace evidence, eyewitness evidence is something that the perpetrator left behind at the scene of the crime. Unlike physical trace evidence, however, a memory trace cannot be observed directly or placed in an evidence container. Nevertheless, if an eyewitness observed the perpetrator commit a crime, then a trace exists inside the witness' head and this trace can be a key to establishing the identity of the perpetrator.

This memory-as-trace-evidence comparison is useful because there are clear parallels between physical trace evidence and memory trace evidence. Like many forms of physical evidence, memory traces can be delicate, and thus easily destroyed or damaged by mishandling. Eyewitness memories can be cross-contaminated (e.g., witnesses interacting and sharing information) just as physical traces can be cross-contaminated (e.g., blood from one area of the crime scene mixed with blood from another area during collection). Memories can decay over time, just as some types of physical evidence can decay over time. How eyewitness memory is tested (e.g., the protocol for conducting a lineup) can influence the reliability of the results just as how one tests physical evidence (e.g., a DNA protocol) can affect the reliability of the results.

Like all comparisons, it is possible to take the memory-as-trace-evidence similarities too far. The comparison serves a useful purpose, however, in that it focuses attention on the fragility of memory and the general idea that the reliability of eyewitness evidence is closely linked to the methods that are used to collect and preserve that evidence. In the case of lineups, the accuracy of a witness's decision should depend solely on that witness's memory for the offender. Injustices can occur readily and frequently when a witness' decision is based on aspects of the procedure that encourage the witness to choose some one in general (as opposed to saying, "I don't know" or "He's not there"), or the suspect in particular (as opposed to other lineup members). Our aim is to provide recommendations that allow investigators to accomplish the goal of obtaining identification evidence with the highest probative value, that is, identification evidence that is maximally informative about the guilt or innocence of the suspect..

The Current State of Affairs
There are currently about 100 documented cases in the U.S. in which a convicted person who has served many years in prison has been exonerated by DNA evidence indicating that someone else committed the crime, all discovered since the early 1990s. Of those 100 cases, over 75% were primarily the result of the mistaken eyewitness identification of the convicted suspect. "Primarily" refers to the fact that in many of those cases other evidence also was presented which seemed to add weight to the overall probability that the suspect was in fact guilty of the crime. However, the other evidence used to augment the eyewitness evidence had generally low probative value (e.g., lack of alibi, matching blood type) and often would not have been collected if the eyewitness had not first identified the suspect. The key to these convictions of innocent persons in over 75% of these cases was the identification testimony of one or more eyewitnesses. Given the (generally) confident testimony of these eyewitnesses at trial, it is not surprising that juries convicted. Research clearly shows that juries accept the identification testimony of confident eyewitnesses. Nor is it surprising that these witnesses were highly confident at trial even though they had mistakenly identified the defendant. Research shows that the confidence of an eyewitness need not be a good indicator of the accuracy of the eyewitness' memory. All of these observations point to the importance of using procedures that help to assure that a mistaken identification does not occur in the first place.

Compounding any estimator-variable problems in a given investigation, system-variable procedures differ greatly across provinces, jurisdictions, police services, and even individual police officers within the same service. Budgets, access to technology, emphases on different crime concerns, and other factors can affect how eyewitness evidence is collected and handled. It is therefore reasonable to assume that not everyone is using the optimal strategy.

Finally, most efforts to address eyewitness errors have been focused on an after-the-fact, case-by-case, approach at the trial phase. At trial, prosecutors argue that the identification was reliable, defense attorneys argue that the identification was not reliable, and occasionally experts testify about factors that affect eyewitness reliability. During this increasingly sophisticated and complex scrutiny of eyewitness evidence at trial, procedures used by law enforcement are being criticized by experts. Many eyewitness researchers recommend a move away from, and ideally the elimination of, expert testimony in court regarding eyewitness evidence problems. The goal is to prevent problems and errors before they occur, through training, sharing knowledge, and a standardized set of recommendations. Such a set of recommendations is presented below.

For many of the reasons described above, in May 1998, at the request of then U.S. Attorney General Janet Reno, the U.S. National Institute of Justice recruited a group of police officers, district attorneys, defense attorneys, and social scientists/eyewitness researchers to address the issue of eyewitness evidence. An initial planning panel of nine representatives from all four disciplines expanded into the 34-member group called the Technical Working Group for Eyewitness Evidence. Seventeen of the 34 members were law enforcement officers, six were district attorneys, five were defense attorneys, and six were social scientists, three of whom are the authors of the current article (see Wells, Malpass, Lindsay, Fisher, Turtle, & Fulero, 2000 for the full story).

A series of meetings held over the course of one year led to the publication of Eyewitness Evidence: A Guide for Law Enforcement which was officially released on October 26 1999 in Washington, D.C. (see Note 2 for details on obtaining the Guide). The Guide is divided into five main sections corresponding to five stages of a criminal investigation specifically concerning eyewitness evidence: (1) managing witnesses at the crime scene, (2) eliciting a description of the offender by searching mug shots or creating a composite likeness, (3) follow-up investigative interviews, (4) field identifications, and (5) lineup identifications. Of course not all of these stages will be important in every case, but at least some are going to be relevant in cases involving eyewitnesses.

Because we helped to create the Guide, and in the interest of promoting a standardized approach to the problem, we will organize our recommendations here using the same layout of its final section: (a) composing lineups, (b) instructing witnesses prior to viewing a lineup, (c) conducting the identification procedure, and (d) recording the identification results. It is important to note, however, that we are not limiting ourselves to the letter of the Guide, and are in fact going beyond the minimum standards described there to include truly "best practice" recommendations. Also, the Guide considers both live- and photo-lineup procedures, but they have most features in common, and photo lineups are far more prevalent than live lineups in Canada, so for the sake of brevity and relevance the procedures and comments provided here assume a photo lineup (see Note 3). For each section, we will begin with a series of steps, followed by a discussion of the rationale for those steps and further comment on their implementation. We are aware that many situations present severe challenges to implementing these recommendations as stated, so our goal is to provide sufficient background logic so that investigators themselves can tailor the recommendations to specific situations.

A. Composing Lineups
1. Include only one suspect in each identification procedure. In cases involving multiple perpetrators and multiple suspects, construct separate lineups for each suspect.
2. Include an appropriate minimum number of fillers (non-suspects) per identification procedure.
3. Select fillers that generally fit the witness' description of the perpetrator. When there is a limited/inadequate description of the perpetrator provided by the witness, or when the description of the perpetrator differs significantly from the suspect, fillers should resemble the suspect in significant features. By definition, any specific feature mentioned by the witness is significant. Thus, if the description states that the criminal had a mole, and the suspect has a mole, a filler without a mole who otherwise closely resembles the suspect is not acceptable. If the suspect does not have a mole (but the witness described a mole) then no lineup member should have a mole. If the original description is particularly vague, common sense should prevail. Thus, if the witness failed to indicate race or sex, these obviously must be matched to the suspect anyway.
4. If multiple photos of the suspect are reasonably available to the investigator, select a photo that resembles the suspect's description or appearance at the time of the incident.
5. Consider that complete uniformity of features is not required. Avoid using fillers that so closely resemble the suspect that even a person familiar with the suspect might find it difficult to distinguish the suspect from the fillers.
6. Create a consistent appearance between the suspect and fillers with respect to any unique or unusual feature (e.g., scars, tattoos) used to describe the perpetrator, by artificially adding or concealing that feature.
7. Place suspects in different positions in each lineup, both across cases and with multiple witnesses in the same case. Alternatively, the suspect or his representative could be allowed to choose his position in the lineup.
8. When showing a new suspect, avoid reusing fillers in lineups shown previously to the same witness.
9. Ensure that no information concerning previous arrest(s) will be visible to the witness.
10. View the lineup to ensure that the suspect does not unduly stand out, either because of a highly distinctive feature absent from other lineup members, or as a better fit to the description than other lineup members (including clothing), or due to differences in pose, background of the photo, lighting differences, etc.
11. Preserve the presentation order of the photo lineup. In addition, the photos themselves should be preserved in their original condition.

This section concerns primarily the issue of how to select who goes into the lineup in addition to the suspect. Item 1 refers to a crucial assumption that by definition a lineup means that there is just a single suspect and that all other members are people known to be innocent of the relevant crime. (Various terms for these additional lineup members are used across police services, including foils, fillers, distracters, stand-ins, shills, and others ­ the term filler will be used here.) In the case of photo-lineups, having a suspect is really the distinguishing feature of a lineup as opposed to a search of arrest-photos ("mug shots") in which there is no specific suspect. The problem with multiple-suspect lineups is that the probability of a possible false identification rises dramatically as the number of suspects in a lineup increases above one, a point originally made by two of the current authors (Wells & Turtle, 1986). The problem can be most easily understood if one considers an extreme situation in which all lineup members are suspects ­ any identification is going to lead to further investigation, as opposed to the single-suspect case in which the identification of any filler provides potentially useful information to police about the quality of the witness's memory, the likelihood that they have the real offender, and possibly the appearance of the real offender (he may look more like the selected filler than like the suspect).

It is important, of course, to conduct a follow-up investigation when an eyewitness makes an identification. The ultimate goal of that investigation is to determine who committed the crime and to rule out innocent suspects. Police sometimes object, therefore, to the notion that an actually innocent person identified in a lineup is at risk of arrest, prosecution, and conviction because they expect that the absence of incriminating evidence and/or the presence of exonerating evidence revealed by the subsequent investigation will rule out the identified person as the actual offender. Granted, if the lineup is conducted early in an investigation, with a low probability that the suspect is the actual offender, then it might not take much to eliminate that person from suspicion, but even in such cases things can go wrong. A mistaken identification is often not a "harmless error" that can be rectified in the ensuing investigation. Coincidental circumstantial evidence, the suspect's criminal history, the legitimate lack of an alibi, and even a false confession are all possible contributors to any initial belief that the suspect is the offender based on a witness's identification from a lineup. The well documented phenomenon of "tunnel vision" (see for example Kaufman, 1998) can emerge to apparently cement the guilt of an innocent person who perhaps would never have even become a suspect without what turned out to be a mistaken identification. We strongly urge, therefore, that all attempts be made to avoid errors at all points of the investigation.

Item 2 in this section recommends using "an appropriate" number of fillers. There is no exact number of fillers that is "correct" or "ideal." Lineup size varies considerably from time to time and country to country from as few as 3 to at least 12. Experience in Canada and Britain with lineups approximately twice the size of the typical American "6-pack" has not led either country to conclude that their lineup size is detrimental to resolving criminal cases. The limited research evidence available indicates that lineup size up to 20 photos does not reduce the likelihood of correct identification. On the other hand, provided that all fillers match the description, the larger the lineup, the less likely an innocent suspect is to be selected by a witness who is guessing. Studies in both the U.S. with 6-person lineups and Britain with 9- and 10-person lineups indicate that approximately 20% of witnesses shown a lineup will identify a filler. This evidence supports concerns that witnesses will identify innocent people from lineups, but also suggests that lineup size in and of itself does not seem to alter the tendency to choose. Overall then, larger lineups should be better and we encourage the use of up to 12-person lineups. It may be prudent to wait for further research before increasing lineup size beyond this level.

Items 3, 4, 5, 6, and 10 in this section all relate to the issue of lineup similarity ­ the degree to which the suspect looks like the rest of the photos and how much they in turn look like each other. Much to their credit, police in the past have often gone to great lengths to ensure that all members of a lineup, including the suspect, look as similar to one another as possible, in an effort to comply with the reasonable and commonly-followed policy that the suspect's photo should not unduly stand out from the rest. Ironically, however, this strategy is not consistent with another reasonable principle that the fair composition of a lineup should facilitate the witness's ability to make an accurate identification of the offender if that offender is in fact in the lineup and the witness has a good memory for him. Selecting photos based on their strong resemblance to the suspect can seriously compromise the ability of a good witness to identify the guilty suspect by creating a lineup where even a person very familiar with the suspect might have difficulty distinguishing him from the others (as the line goes, "His own mother couldn't recognize him among those guys"). It is often counterintuitive, then, for police officers to hear from us that lineup distracters must merely match the description of the offender as given by the witness viewing that lineup, as long as the policy that the suspect does not stand out is upheld. More on this point is available in Wells, Rydell, and Seelau (1993), but suffice it to say here that the goal is actually to create some beneficial variety among the photos, without making the lineup biased against the suspect. We have two further recommendations to accomplish this goal.

Our first recommendation for achieving a desirable level of match of fillers to the suspect involves an "iterative" strategy for selecting lineup fillers, which works as follows: (1) Conduct the initial search of your data base for appropriate photos using search criteria based on the witness' description of the offender and not the typically more detailed information available in the suspect's record in the system. This will ensure that the initial pool of appropriate photos is not too similar in appearance before any further selection is started. (2) Select the first filler for the lineup to be as highly similar to the suspect as you want or are used to using. (3) Put the photo of the suspect out of sight. (4) Now select the second filler to be as highly similar as you want to the first filler you just selected. (5) Put the photo of the first filler out of sight. (6) Select the third filler to be as highly similar as you want to the second filler you just selected. (7) Continue in this manner until you have selected one more filler than required. (8) Discard the first filler selected and use the remaining fillers in the lineup. The resulting lineup contains no fillers selected explicitly because of their similarity to the suspect, but all should match the suspect's general appearance. (9) Examine the final lineup including the suspect to determine if any filler can be excluded based on the description provided by the witness. If a filler can be eliminated, replace that person with a further choice (this should rarely be necessary).

Our second recommendation involves the rule of thumb that a person who has never seen the offender before should not be able to identify the suspect in the lineup other than by guessing. Suppose, for example, that the offender in a particular case was described by the witness as being a white male in his mid 20s with curly blonde hair and a moustache. Suppose further that you have constructed a lineup around a person who matches that description and who, likely for additional reasons, is the person suspected of committing the crime. You might have used our iterative strategy described above, or another strategy. Now provide the description of the offender to 10 people totally unrelated to the case (colleagues, administrative staff, cleaning staff, etc.), show them the lineup, and ask them which person they think is the offender. If, say, 8 out of 10 select your suspect, then you have good reason to believe that the lineup is biased against that suspect because these "mock witnesses" who have never seen the offender before can "identify" him from the lineup! On the other hand, if these "mock witnesses" collectively say something like, "How the hell should I know?! All of these guys look like the person you described", their choices will spread across several photos in the lineup, and you have good reason to believe that your lineup is not biased against the suspect. For more on this "mock witness" procedure, refer to articles in a special issue of Applied Cognitive Psychology dedicated to the topic (e.g., Malpass & Lindsay, 1999).

A final point worth making about selecting lineup fillers is that all of the procedures described above are based on the description provided by the witness. If there is more than one witness, it will often be the case that their descriptions differ. A lineup that is unbiased for one witness may be biased for another witness in the same case. If witnesses' descriptions of the perpetrator are sufficiently similar, it may be reasonable to use the same lineup for all witnesses. However, if the witnesses provide significantly different details, it is appropriate to use a different lineup for each witness with each lineup constructed specifically based on the description provided by the witness who will be shown the lineup.

B. Instructions to a Witness Prior to Viewing the Lineup
(Note that these are generic instructions ­ additional points are added in the next section for the specific type of lineup procedure we recommend there.)
1. Instruct the witness that he/she will be asked to view a set of photographs.
2. Instruct the witness that it is just as important to clear innocent persons from suspicion as to identify guilty parties.
3. Instruct the witness that individuals depicted in lineup photos may not appear exactly as they did on the date of the incident because features such as head and facial hair are subject to change.
4. Instruct the witness that the person who committed the crime may or may not be in the set of photographs being presented.
5. Assure the witness that regardless of whether an identification is made, the police will continue to investigate the incident.
6. Instruct the witness that the procedure requires the investigator to ask the witness to state, in his/her own words, how certain he/she is of any identification.

Item 4 is probably the most important in this subsection, and Item 6 is definitely the most controversial (in close competition perhaps with Item 2). Item 4 is essentially a "None of the above" alternative that makes explicit the possibility that the police suspect is not in fact the offender. It is of course hoped by everyone (except the offender) that the suspect is indeed the offender, and it is likely to be the case in the majority of lineups, but of course the whole rationale behind the procedure is to test that very hypothesis. This instruction, therefore, counters a strong tendency for people to use a "relative judgment strategy" and choose the person who merely looks most like the offender they saw, based on the assumption that the police would not bother with the procedure if they did not have the guilty party in the lineup. Other elements of our recommended procedures discussed in the next section also are designed to reduce witnesses' use of the relative judgment strategy.

The request for a statement of certainty (Item 6) is one that both the police and district attorney representatives had great difficulty accepting in putting together the U.S. Guide. There is a crucial reason, however, for its inclusion both in the Guide and in these best practice recommendations. The arguments for its exclusion are often based on two unfounded assumptions. Our reason for including the instruction is based on common sense, common practice, and now a large body of evidence showing that people's confidence in their lineup selection can be inflated as a result of what transpires after the procedure has ended (see for example Bradfield, Wells, & Olson, 2002; Wells & Bradfield, 1998, 1999). Many cues, both implicit and explicit, can be present that can indicate to a witness that the police believe he or she has made the "correct" choice (that is, identifying the suspect). It is of course beneficial for the prosecution to have a witness appear as confident as possible when asked about the lineup choice in court (e.g., "How confident are you that the man you selected from the lineup is in fact the person you saw commit the crime?"). But if that confidence is proffered as being indicative of how the witness truly felt at the time of the identification, we argue that there must be an actual measure taken right after the choice is made, before the risks of "confidence inflation" are encountered. We have two other recommendations (blind lineups and videotaping) coming up, however, that may render this instruction unnecessary or at least certainly less controversial, but the rationale behind it is still strong.

The first of the two arguments for excluding this instruction concerns the often perceived assumption that we think police are intentionally trying to make witnesses feel more confident than they really are. This is not the case. The same issue arises in the next section on how to actually conduct the lineup procedure, and we will address it there more fully. Suffice it to say here that we are more concerned with subtle and unintentional yet insidious behaviour than we are with blatant examples of attempted influence or misconduct. Our recommendations are based on human behavioural tendencies, including the human police officer's reaction to the witness's lineup choice, and not on any assumption of ulterior motives on the part of the police. Officers motivated to distort or create evidence will not be prevented from doing so by these or any other guidelines for conducting lineups. We believe such behaviour is very rare and present these suggestions fully expecting that the vast majority of officers share our concern for obtaining the best and most object evidence possible.

The second argument for excluding the confidence-statement instruction concerns the issue of a scaled response. There is perhaps legitimate concern that requiring a numerical value (e.g., "How confident are you on a scale from 1 to 10?") might result in an arbitrarily lower value that is dependent on an individual's threshold for using the scale's upper values. Neither the Guide nor we, however, are recommending that witness confidence be assessed on any kind of scale. The recommendation that the procedure require the investigator to ask the witness to state, in his/her own words, how certain he or she is of any identification is meant to elicit any statement that describes the witness's confidence at that time. If the witness responds with, "Do you mean on a scale from 1 to 10?", the officer can reply, "Not necessarily ­ whichever way best describes your confidence in your choice".

C. Conducting the Identification Procedure for a Blind, Sequential Photo Lineup
The crucial elements of conducting a "blind, sequential" lineup are included in this section. Before we get to the specific procedures, however, it is important to describe these terms and the rationale behind them. "Blind" refers to the fact that the officer conducting the identification procedure is either (a) unaware of which lineup member is the suspect or (b) not able to see which picture the witness is viewing at any given time (more on these alternatives follows below). As we mentioned in the previous section, our experience has been that officers are often quick to react to this type of recommendation because they believe it assumes they are intentionally trying to influence the procedure and its outcome. Again, this is not the case. Recommendations of this sort merely recognize the human element in this procedure ­ that the most well-intentioned person can unintentionally convey information and expectations to the witness that can lead that witness to make a particular decision. This is not a "psychic" phenomenon, but rather a mundane part of human nature. We have borrowed the term "blind" from its decades of use in scientific experiments and clinical drug trials. Scientists and medical doctors are rarely accused of intentionally influencing the outcomes of their experiments and drug tests, but it is accepted as fact that they can have an unintentional influence on the results if they have too much information about a particular data point in an experiment or a patient in a drug trial. This is such a real possibility that nearly all clinical drug trials involve a "placebo" or "control" group in order to create a "double-blind" situation, in which both the patient taking the pills and the doctor assessing the patient's health don't know if the pills are the actual new drug or an inert substance.

So we are merely recommending that police recognize that they have the same potential to unintentionally influence the outcome of an identification procedure as a doctor does to inadvertently "see" improvement in a patient if he or she knows that the patient has been taking the actual drug. The classic example from the lineup situation goes something like this: The officer knows that Number 3 is the suspect. The female witness pauses on Number 2 and says, "This person looks a lot like the guy I saw", to which the officer responds in a very professional and apparently objective manner, "Take your time ma'am", to which the witness replies, "Okay". The witness next pauses on Number 3 and says, "This person looks a lot like the guy I saw too", to which the officer responds in a very professional and apparently objective manner, "What is it about Number 3 that you recognize, ma'am"? It doesn't look like much, but if a witness does not have a good memory for the actual offender and/or the offender is in fact not in the lineup, then such a procedure can functionally direct the witness to choose the suspect without either the witness or the officer believing that's what happened. We think of it as like the "Ouji board" phenomenon, rather than dishonest behaviour on the part of the police.

The other element of the "blind, sequential" lineup has to do with how the photos are actually presented to the witness. The traditional procedure is to present all of the photos at once, or "simultaneously", in an array of 6 to 12 photos. The problem, however, is that such a procedure can encourage the witness to use the "relative-judgment" strategy we discussed earlier in the context of the pre-lineup instruction, "The actual offender may or may not be in the photos". Again, common sense, research results, and officers' actual experience all show that it is not uncommon for a witness to explicitly employ a sort of process of elimination, whereby the person who looks most like the offender is identified. An alternative identification technique, sequential lineup presentation, goes a long way toward solving this problem and several others. The idea was first introduced by two of the current authors over 15 years ago (Lindsay and Wells, 1985), and Lindsay and his colleagues have gone on to conduct numerous studies with thousands of participants to hone the technique and demonstrate its effectiveness (see for example Lindsay, Lea, Nosworthy, Fulford, Hector, LeVan, & Seabrook, 1991).

The sequential technique requires a combination of several simple procedures: (a) each lineup member or photograph is presented individually, (b) witnesses are not told how many people are in the lineup, and (c) witnesses are informed that they may take as long as they wish looking at each person but that once they have decided, that their decision is final ­ that is, they will not be allowed to go back over the lineup again. As each person or photograph is presented, the witness is required to make a decision of whether or not it is the perpetrator. A rapidly growing body of research shows that sequential lineup presentation makes it extremely difficult to use a relative judgment strategy. As a result false identifications by eyewitnesses occur at a dramatically reduced rate while, fortunately, the rate of accurate identifications is not reduced significantly (the tradeoff between reducing false identifications and losing accurate ones is discussed in more detail below ).
Here now are the step-by-step procedural recommendations for conducting a blind, sequential lineup, followed by more discussion to facilitate their implementation and deal with potential problems.

1. Provide viewing instructions to the witness as outlined in Section B, as well as the following:
a. Individual photographs will be viewed one at a time.
b. The photos are in random order.
c.The officer presenting the pictures does not know which person is the suspect.
OR
c. The officer does not know which position the suspect's photo is in and won't be able to see which photo you are viewing (much more on this below).
d. Take as much time as needed in making a decision about each photo before moving on to the next one, but a clear decision must be made and stated before the next photo will be shown.
e. Our departmental policy is that all photos will be shown, even if an identification is made.
2. Confirm that the witness understands the nature of the sequential procedure.
3. Begin with all photographs out of the view of the witness.
4. Instruct all those present at the lineup not to suggest in any way the position or identity of the suspect in the lineup.
5. Avoid saying anything to the witness that may influence the witness's selection.
6. Present each photo to the witness separately, removing the one previously seen before the next photo is shown.
7. Ensure that a clear decision is made indicating whether or not the witness believes the person being examined is the criminal before exposing the witness to the next lineup member.
8. If an identification is made, avoid reporting to the witness any information regarding the individual he/she has selected prior to obtaining the witness' statement of certainty.

There are at least two ways to conduct a blind lineup. First, the officer conducting the procedure does not know who the suspect is. This can be accomplished by having a person other than the investigating officer handle the task. Under these conditions the officer may present the photos to the witness without concern that the officer can see the photos when presented. Frequently, however, there will not be an officer unaware of the identity of the suspect (e.g., in a big case and/or a small police service) or personnel limitations may prevent use of an officer unaware of the identity of the suspect (e.g., the officer conducting the lineup may have to be paid court time for testifying at trial).

The second possible strategy for conducting a blind lineup can be easier to implement, depending on the nature of the case, the size of the police service, and the available technology ­ the point is to have the officer "blind" to the photos that the witness is viewing. An advantage to this alternative is that the investigating officer familiar with the case has the opportunity to view the witness' behaviour, which could be valuable, say, in the case of a witness's emotional reaction to what turns out to be the suspect's photo. A "low-tech" version of this technique is to place each picture in an envelope, shuffle the envelopes, number the envelopes, then hand them to the witness one at a time. The witness will be instructed to take out the photo, announce (or write) the decision regarding the photo, replace the photo in the envelope, and return it to the officer. Once an envelope is returned to the officer, the next envelope is passed to the witness. The officer is careful that she or he is not in a position to see the pictures while the witness is examining them and does not examine the photos when they are returned. A "high tech" version of this alternative is to use computer presentation of the lineup. If the order of presentation of the pictures is determined by the computer or pre-set by another officer prior to the lineup being shown to the witness, the officer would be blind so long as he or she was in a position that did not allow him or her to see the screen. It is our experience that police services often do not exploit the full functionality of the typically expensive and sophisticated software they use to store their arrest photos when it comes to constructing and presenting actual lineups. Here is one way to take fuller advantage.

There are potential problems to using the sequential technique versus the more traditional simultaneous presentation of photos. Opportunities to examine the lineup members a second time, for example, are dangerous when no one was selected the first time through the faces. It may be necessary to allow a witness to view the pictures a second time, and it's possible that the witness will subsequently identify someone they passed on the first time through. In this situation we recommend that the results of the procedure be well documented and "go to weight" in terms of their value in the case and any subsequent trial. The officer can in good conscience say, for example, "The witness paused on the Number 5 for 10 seconds during the sequential presentation of the photos, but finally said it wasn't him and moved on to the other photos, on which she spent no more than 2 seconds each. After viewing the last photo in the series the witness said, 'I knew it was that guycan I go back to Number 5?!' At this point I allowed her to view Number 5 again and she said 'I should have picked him the first time through, but his hair was different when I saw him so I was hesitant to pick him without seeing the others.' It was later revealed that Number 5 was the suspect in the case". The officer is not claiming that the outcome followed from the strict administration of a formal sequential-lineup presentation, but there is still information there, the value of which can be weighed in subsequent phases of the investigation and possibly at trial. However, this evidence is clearly weaker than an identification made upon first exposure to the suspect.

The same problem does not apply when an identification is made and the witness requests an opportunity to examine the remaining lineup members. Data indicate that once eyewitnesses have selected the guilty person, they are unlikely to change their mind and choose someone else. Denying the witness such an opportunity could be perceived by the defense and the courts as biased because witnesses are prevented from impeaching their identification by selecting another person. The decision to show any remaining photos is a policy issue, and should not depend on the outcome of a particular lineup presentation. We recommend in Item 1e, therefore, that witnesses be told that all photos will be shown even if an identification is made, and that the procedure be followed in all appropriate cases.

Very recent data indicate that in four specific situations sequential lineups may be no better or even worse than the traditional simultaneous lineup. First, if the witnesses are children (10 or under) they may be confused by the sequential lineup procedure, resulting in lower correct identification rates than from traditional lineups, multiple selections from the sequential lineup, and/or equal false identification rates from both techniques. Second, if the suspect does not match the description originally given by the witness on a central detail (e.g., the criminal was bearded but the suspect is clean shaven), sequential lineups may result in substantially lower correct identification rates than traditional lineups unless the witness has been instructed prior to the lineup that all lineup members will share the change in appearance (e.g., be clean shaven). Third, if multiple perpetrators were involved in the crime and more than one suspect is to be shown to the witness(es), it is not clear how a sequential procedure should be used and traditional methods have not been shown to be inferior in such cases. On the other hand, all methods tested to date are prone to low rates of correct identification and high rates of false identification in multiple perpetrator situations. And fourth, cross-race identifications do not appear to benefit from sequential presentation of lineups. None of these potential problems has been investigated sufficiently at this time to draw conclusions other than that the sequential lineup has not been demonstrated to show its superiority under these conditions, and some data exist suggesting that there may be some disadvantage to using the procedure under these conditions. Until more and better data are available, we do not recommend using sequential lineups in these particular situations.

Perhaps the hottest question concerns the issue of whether there is a reduction of accurate identifications resulting from the sequential lineup. Although the reduction of mistaken identifications from the sequential lineup is well established and robust, some have been surprised to encounter the idea that accurate identification rates can also be lower with the sequential lineup than with the simultaneous lineup. Some loss of accurate identifications resulting from the sequential lineup should not be surprising, however, given the psychological mechanisms involved. This is because some of the "accurate identifications" that come from the simultaneous lineup are merely the result of the relative-judgment strategy we discussed earlier. In other words, some witnesses who have rather weak memories nevertheless pick out the suspect because they are simply picking the person who looks most like the culprit compared to the other lineup members. These weak-memory witnesses would not, however, be able to identify the culprit from a sequential lineup because they don't have a good enough memory to do anything but make the shallow relative decision. In a sense, these witnesses are guessing and happen to guess "correctly" ­ that is, by selecting the suspect. The sequential lineup largely eliminates this type of guessing and, therefore, there is some loss in identifications of the suspect.

But the sequential-superiority effect can be defined as the ability of the sequential lineup to produce a higher overall ratio of accurate to mistaken identifications of suspects, in spite of some loss in accurate identifications. The most complete discussion of this issue is a meta-analysis of the 9 published and 13 unpublished papers on simultaneous versus sequential lineups, based on data from 4,125 research participants (Steblay, Deisert, Fulero, & Lindsay, 2001). The results of the Steblay et al. analysis reveal that correct identification rates of the culprit when the culprit is in the lineup are 50% for the simultaneous presentation and 35% for the sequential. So, the sequential yields only 70% of the "hits" that the simultaneous does. (The actual differences are likely to be much smaller than this in actual cases for reasons outlined in the Steblay et al article; however, we will assume this magnitude of difference in hit rates so as to deal with the "worst case scenario".) Mistaken identification of an innocent "stand in" for the culprit is 27% for the simultaneous and 9% for the sequential. So, the sequential yields only 33% of the "false alarms" that the simultaneous yields. Overall, then, the hit-to-false-alarm ratio for identifications of the suspect for simultaneous lineups is 50:27 (or slightly under 2.0) compared to 35:9 (or nearly 4.0) for sequential lineups in these studies. This means that the odds of an identification of the suspect being accurate are approximately doubled by the use of the sequential lineup. Although some policy makers have expressed concern over the loss in accurate identifications that can result from the sequential lineup, we argue that its advantage lies in generating evidence that is more diagnostic of the suspect's guilt.

D. Recording Identification Results
1. Record both identification and nonidentification results in writing, including the witness' own words regarding how sure he/she is and spontaneous comments of any kind..
2. Ensure results are signed and dated by the witness.
3.Ensure that no materials indicating previous identification results are visible to the witness.
4.Ensure that the witness does not write on or mark any materials that will be used in other identification procedures.
5. Document in writing the photo lineup procedures, including:
a.Identification information and sources of all photos used.
b.Names of all persons present at the photo lineup.
c. Date and time of the identification procedure.
6. Preserve a copy of the lineup by photo or video. This documentation should be of a quality that represents the lineup clearly and fairly. Photo documentation can be either the group or each individual.
7. Instruct the witness not to discuss the identification procedure or its results with other witnesses involved in the case and discourage contact with the media.

Item 1 is an important point, even though it is not necessarily obvious that a witness who selects no one from the lineup might be providing useful information to the investigator. Two of the current authors first discussed this issue over 20 years ago (Wells & Lindsay, 1980) and refined more recently in Wells and Olson (2002). In short, if one or more witnesses does fail to choose anyone and/or they identify fillers from the lineup, police should re-evaluate the likelihood that their suspect is in fact the offender with as much vigor as they would pursue the suspect if an identification were made. Finally, in addition to the possible benefit to solving the case, it could appear to the defense that only recording and admitting as evidence the results of lineups that lead to an identification (most likely of the accused) is akin to withholding evidence.
Item 6 can refer to two separate issues ­ on the one hand, merely retaining a copy of the lineup and/or enough information to recreate it from a database of photos, but on the other hand the possibility of recording the actual procedure itself by either audio- or videotape recording. We hesitate to recommend the latter idea as strenuously as our others because we realize that it poses significant implementation problems on a large scale basis (although see Kassin, 1998). But much can be gained in some cases; for example, when police have good reason to believe that the identification may be challenged later (e.g., by a defense counsel who is known to scrutinize identification procedures at trial), or where a witness is expected to express an emotional reaction to the photo of the suspect that police predict might be a persuasive indicator of accuracy to the defense, judge, and/or jury (e.g., a younger witness who is not very verbal, but is thought to have got a good look at the offender).

A final issue in this section is the use of identification forms, which are often used but which may be in need of revising. Eyewitness researchers use forms with very specific features when they conduct identification procedures, especially when using the sequential presentation technique. We recommend a form that requires the witness to provide an explicit response for each photo; for example, "Is Number 1 the person you saw? Circle YES or NO. Is Number 2 the person you saw? Circle YES or NO." And so on. Note that researchers usually provide more lines than the number of lineup members in order to reduce pressure on the witness to choose as the end of the sequence approaches (thus a form for a 12-person lineup could go up to, "Is Number 20 the person you saw? Circle YES or NO." so that when the person is at what is actually the final photo in position 12 they are not more likely to choose it just because it is at the end of the series). And finally, of course, there needs to be the question, "How certain are you that the decisions indicated above are correct?", with some blank lines or empty space afterward.
Using this method of recording witness responses makes it clear to the witness that only a yes or no response is acceptable and may prevent some of the attempts to circumvent the sequential procedure described above. An alternative phrasing of the question could be used to avoid situations where the witness feels clearly that she is not able to positively identify the person but is equally unsure that the person is not the criminal. To avoid this problem, the question could be phrased, "Can you positively identify this person as the person you saw?" A "no" answer to this question does not indicate that the witness believes that the person is not the criminal, only that the witness is unable to positively confirm that the person is the criminal. As we mentioned above, a computer-based presentation of the lineup should be seriously considered; in this case, it can facilitate the completion of a form as the witness clicks "YES" or "NO" buttons, as well as provide information such as the time spent on each photo, and a reliable record of the photos used and the order in which they were presented. The form could be printed off and kept as an official record of the procedure.

We want to stress again that these recommendations demonstrate a preference for avoiding problems with eyewitness evidence before they occur. We are not suggesting that experts be consulted more often than they are now, and ideally we see them being consulted less as police implement procedures that are less vulnerable to challenges from the defense and less likely to elicit inaccurate identifications that need to be reviewed later. We argue that four basic points will facilitate this change. First, we recommend that police accept the basic logic behind these recommendations ­ that a witness's identification decision should be based solely on his or her memory for the offender, and not influenced by features of the lineup procedure that encourage decisions to go one way or another. Second, we recommend that officers be trained to construct their own lineups, as opposed to having them done by dedicated officers or civilians who may not appreciate the particulars of a given case. We recommend that appropriate training, based on these recommendations, their logic, and the corresponding research literature, be undertaken so that police investigators can react accordingly when faced with a situation that is not specifically addressed in the steps as they are laid out here. We leave it to individual police services to decide if they want to implement the blind-lineup recommendation using additional officers or by having the investigating officer do it following our suggested strategies. Third, and to facilitate the previous point, we recommend that computers be exploited further to construct, present, and record lineups. It is our contention that computers will eventually be the sole mechanism by which these tasks are done, so we encourage police services to make that happen sooner rather than later and reap the accompanying benefits. And fourth, it is vitally important to ensure that these recommendations are effective in practice on a large scale, so further research should be conducted to evaluate outcome variables such as arrest rates, police satisfaction, conviction rates, and court challenges in real cases involving eyewitness evidence and use of the procedures we have recommended here.
Our hope is that law enforcement in particular and the administration of justice in general will benefit from these recommendations. Their goal is to maximize the quantity, accuracy, and value of eyewitness evidence, while minimizing errors that can potentially waste time in an investigation, cause unnecessary hardship for an innocent person, and of course leave the actual offender still at large. Through continued research, evaluation, development, and feedback it is also hoped that the recommendations will evolve to become even more useful.

Bradfield, A.L., Wells, G.L., & Olson, E. A. (2002). The damaging effect of confirming feedback on the relation between eyewitness certainty and identification accuracy. Journal of Applied Psychology, 87, 112-120.

Cutler, B.L., Fisher, R.P., & Chicvara, C.L., (1989). Eyewitness identification from live versus videotaped lineups. Forensic Reports, 2, 93-106.

Kassin, S. M. (1998). Eyewitness identification procedures: The fifth rule. Law & Human Behavior, 22, 649-653.

Kaufman, F. (1998). The Commission of Proceedings Involving Guy Paul Morin. Toronto, ON: Queen's Printer.

Kebbell, M.R. (2000). The law concerning the conduct of lineups in England and Wales: How well does it satisfy the recommendations of the American Psychology-Law Society? Law & Human Behavior, 24, 309-315

Lindsay, R. C. L., & Wells, G. L. (1985). Improving eyewitness identification from lineups: Simultaneous versus sequential lineup presentations. Journal of Applied Psychology, 70, 556-564.

Lindsay, R.C.L., Lea, J.A., Nosworthy, G.J., Fulford, J.A., Hector, J., LeVan, V., & Seabrook, C. (1991). Biased lineups: Sequential presentation reduces the problem. Journal of Applied Psychology, 76, 796-802.

Lindsay, R. C. L., & Wells, G. L. (1985). Improving eyewitness identification from lineups: Simultaneous versus sequential lineup presentations. Journal of Applied Psychology, 70, 556-564.

Malpass, R.S. & Lindsay, R.C.L. (1999). Measuring line-up fairness. Applied Cognitive Psychology, 13, S1-S7.

R. v. McGuiness, Ballantyne & Ballantyne, [1997] B.J. No.1695 (B.C.S.C.) (Q.L.).

R. v. Sophonow [1985] Manitoba Court of Appeal. Recommendations of the subsequent inquiry available at http://www.gov.mb.ca/justice/sophonow/recommendations/english.html

Steblay, N. M., Deisert, J., Fulero, S. & Lindsay, R.C.L. (2001). Eyewitness accuracy rates in sequential and simultaneous lineup presentations: A meta-analytic comparison. Law and Human Behaviour, 25, 459-474.

U.S. National Institute of Justice (1996). Convicted by juries, exonerated by science: Case studies in the use of DNA evidence to establish innocence after trial. Washington, D.C.: NCJ 161258.

U.S. National Institute of Justice (1999). Eyewitness evidence: A guide for law enforcement. Washington, D.C.: NCJ 178240.

Wells, G. L. (1978). Applied eyewitness-testimony research: System variables and estimator variables. Journal of Personality and Social Psychology, 12, 1546-1557.

Wells, G. L. (1995). Scientific study of witness memory: Implications for public policy and law. Psychology, Public Policy, and Law, 1, 726-731.

Wells, G. L., & Lindsay, R. C. L. (1980). On estimating the diagnosticity of eyewitness nonidentifications. Psychological Bulletin, 88, 776-784.

Wells, G. L., & Bradfield, A. L. (1998). "Good, you identified the suspect": Feedback to eyewitnesses distorts their reports of the witnessing experience. Journal of Applied Psychology, 83.

Wells, G. L. & Bradfield, A. L. (1999). Distortions in eyewitnesses' recollections: Can the postidentification feedback effect be moderated? Psychological Science, 10, 138-144.

Wells, G.L., Malpass, R.S., Lindsay, R.C.L., Turtle, J.W., and Fulero, S.M. (2000). From the lab to the police station: A successful application of eyewitness research. American Psychologist, 55, 581-598.

Wells, G.L., Rydell, S. M., & Seelau, E.P. (1993). On the selection of distractors for eyewitness lineups. Journal of Applied Psychology, 78, 835-844.

Wells, G.L., & Turtle, J.W. (1986). Eyewitness identification: The importance of lineup models. Psychological Bulletin, 99, 320-329.

Wells, G. L. & Olson, E. (2002). Eyewitness identification: Information gain from incriminating and exonerating behaviors. Journal of Experimental Psychology: Applied.

Note 1. Similar recommendations are discussed in other sources. Interested readers are encouraged to review these recent articles for further information:
Levi, A. M & Lindsay, R C L (2001). Lineup and photo spread procedures: Issues concerning policy recommendations. Psychology, Public Policy, & Law, 7, 776-790.
Lindsay, R.C.L. (1999). Eyewitness evidence. In G. Chayko and E. Gulliver (Eds.), Forensic Evidence in Canada, 2nd ed. (pp. 53-83). Aurora, ON: Canada Law Book.
Wells, G. L. (2001). Police lineups: Data, theory, and policy. Psychology, Public Policy, & Law, 7, 791-801.
Wells, G. L., Small, M, Penrod, S., Malpass, R. S, Fulero, S. M., Brimacombe, C.A.E. (1998). Eyewitness identification procedures: Recommendations for lineups and photospreads. Law & Human Behavior., 22, 603-647.

Note 2. The book Eyewitness Evidence: A Guide for Law Enforcement is a U.S. National Institute of Justice publication, # NCJ 178240. It can be obtained free of charge by writing to the following address: U.S. Department of Justice, Office of Justice Programs, National Institute of Justice, Washington, D.C. 20531. The Guide can also be ordered by calling 301-519-5500, or downloaded as an ASCII text file or Adobe Acrobat (PDF) file from http://www.ojp.usdoj.gov/nij/pubs-sum/178240.htm

Note 3. Despite general assumptions that live lineups must be superior, data suggest that differences in correct and false identification rates are similar using live lineups and photo arrays (Cutler, Fisher, & Chicvara, 1989). Photo arrays have an additional advantage because it is easier to construct fair photo arrays due to the availability of large photo pools (mug shots). In addition, live lineups can be thwarted by suspects acting out in ways that draw attention to themselves and make interpretation of their identification ambiguous.