Drawing of a courtroom scene

An eyewitness’s identification of an accused defendant often provides some of the most dramatic and powerful evidence in a criminal case. “Do you see in this courtroom the person you saw fire the fatal shot?” asks the prosecutor. “Yes,” says the eyewitness, pointing to the defendant, adding for good measure, “I will never forget his face.”

But in fact the eyewitness is frequently wrong: inaccurate eyewitness identifications appear to be the single greatest contributor to wrongful convictions. For example, they were introduced as evidence in over 70 percent of the more than 360 cases that the Innocence Project, using DNA analysis, later proved were wrongful convictions. Nearly a third of these cases, moreover, involved multiple misidentifications of the defendant. By comparison, the next-most-frequent contributor to wrongful convictions, misleading testimony by forensic “experts,” was present in 45 percent of these cases, and the third-most-frequent factor, false confession, was present in about 30 percent of them.

While some eyewitnesses have had prior contact with the person they identify as the perpetrator of a crime (as when a neighbor sees a husband abusing his wife), many have had none and only see the defendant once, when they witness the crime. But in some respects this makes their testimony stronger, for they have no motive to lie. The defendant was a complete stranger to them, and they simply had the misfortune to have been a passer-by or, worse, a victim. In either case, the encounter was not something they were likely to forget—and the jury generally finds their testimony believable.

Why are eyewitnesses so often wrong? Improper police practices sometimes play a part, as when a police officer conducting a line-up urges the eyewitness to “take a good look at number 3,” or when the eyewitness only tentatively identifies the person in the line-up that the officer suspects is the culprit and the officer says, “Good work.” But the chief causes of inaccurate eyewitness identifications are shortcomings inherent in human perception and memory that cannot be eliminated easily, if at all. Some of these are obvious. The ability of an eyewitness to perceive the face of a culprit will be affected by lighting, by distance and angle, by the acuity of the eyewitness’s eyesight, by the amount of time the eyewitness looked at the culprit, and by distractions such as a gun. Similarly, memories tend to fade over time, which may affect how accurately an eyewitness can remember a face seen many hours, days, or even weeks earlier.

Considerable research indicates, however, that many people overrate their ability to perceive and remember faces they saw only once, and that what they remember mostly relates to some general characteristic, such as that the culprit was square-jawed or had a mustache.1 The research also shows that there are many other factors that can influence and distort an eyewitness’s perceptions and recollections. For instance, careful studies going back as far as the 1980s have demonstrated what is now called the “other-race effect”: “eyewitnesses are less likely to misidentify someone of their own race than they are to misidentify someone of another race.”2 While theories vary as to why this is so, it is now pretty much beyond dispute.

Another less-than-obvious factor is the result of the well-known tendency of memories to merge over time in order to “fill in the gaps.” An eyewitness who does not know the identity of the perpetrator, say, will thus often be asked by police to view a line-up or photo array that includes one or more possible suspects, to see if the eyewitness can pick out any of them as the culprit. At the time of this viewing, the eyewitness may have only a somewhat blurred memory of the crime, but will typically study the line-up or photographs with care before making a selection (if any). By the time the eyewitness testifies at trial, however, the rough memory of the perpetrator from the actual time of the crime will often have merged with the memory formed from the much more careful scrutiny of the line-up or photographs, so that the eyewitness honestly thinks he remembers a particular detail, like a scar on the defendant’s face, from the time of the crime, even though his perception of that detail came from his viewing of the line-up or photographs. More generally, as summarized in a recent article in The New York Times, “memories formed during a traumatic event become an amalgamation of a person’s understanding of the world, the people around them and the snippets they were able to encode into their brain.”3

Common assumptions can also potentially distort an eyewitness identification. For example, even if a well-trained police officer makes a point of telling an eyewitness not to assume that any of the people being viewed in a line-up or photo array is a suspect, most eyewitnesses will assume that one or more are indeed included, and this will increase the likelihood that the eyewitness will make an identification.


To be sure, not all police line-ups and photo arrays are conducted so carefully, and many of the legal developments of the last few decades regarding eyewitness identifications have focused on requiring less suggestive police procedures, such as having a line-up or photo array conducted by an officer not involved in the investigation of the crime. The purpose of this reform is to eliminate the possibility that the officer will suggest, if only through body language, that the eyewitness should make a particular selection. But such reforms, though salutary, are largely irrelevant to solving the more basic problems of human perception and memory that appear to be the main cause of so many false identifications.

Some of these misidentifications have been astonishing. Consider the following three cases.

In 1984 Kirk Bloodsworth was convicted and sentenced to death for the rape and murder of a nine-year-old girl in Baltimore. While no physical or circumstantial evidence linked him to the crime, no fewer than five eyewitnesses placed him with the victim or at the scene of the crime. At the time, DNA testing had not yet made its way into the criminal justice system; the first US case involving its use by prosecutors was in 1988, and defense lawyers did not begin to use it until a few years later. Finally, in 1993, DNA analysis of the semen extracted from the girl’s underwear showed that the culprit was not Bloodsworth but someone else, who eventually confessed. Thankfully, Bloodsworth had not been executed, and he was set free that year.

Also in 1984, a college student named Jennifer Thompson was raped in Burlington, North Carolina. When shown an array of six photos, she tentatively identified Ronald Cotton as her assailant, initially stating, “I think this is the guy.” At the time of trial, however, Thompson testified that she was “absolutely sure” that Cotton was the man who had raped her. Cotton was convicted and sentenced to life imprisonment. Over a decade later, DNA testing of the semen taken from her vagina right after the crime proved to be that of another man, who was then charged, and Cotton was set free.

In 1974 James Bain was convicted in Florida of raping a nine-year-old boy. Although this was long before DNA testing was available, blood found in the semen taken from the boy’s underwear was type B, and Bain’s blood was type AB. Nevertheless, the jury convicted him, chiefly on the basis of the boy’s identification of Bain, which he had consistently provided from the initial photo array through the trail. Years later, after DNA testing became available, Bain filed handwritten motions in court four times asking for DNA testing of the semen, but all four were denied. Finally, however, after Bain obtained assistance from a lawyer and the Innocence Project of Florida, DNA testing was granted—and it completely exonerated him. He was released in 2009, having served thirty-five years for a crime he did not commit.

In each of these cases, had it not been for DNA testing, the defendants would still be in prison, or dead. But DNA samples are either unavailable or irrelevant to the investigation and prosecution of most crimes. It may reasonably be inferred, therefore, that numerous defendants currently imprisoned were wrongly convicted on the basis of inaccurate eyewitness testimony. And it may be further assumed that while some of these inaccurate identifications may have been the product of suggestive police procedures, many more were the result of shortcomings in perception and memory that are endemic to the human species.

How should the legal system deal with this seemingly intractable problem? The fact that eyewitness identifications are frequently unreliable was recognized by the Supreme Court as early as 1967 in cases like United States v. Wade, Gilbert v. California, and Stovall v. Denno, which focused on the need to have defense counsel present at line-ups in order to avoid unfairness. It was not until a decade later, in 1977, that the Supreme Court, in Manson v. Brathwaite, addressed the issue of whether eyewitness identifications might be the product of overly suggestive police techniques, such as, in Manson, showing the eyewitness a single photograph. The emphasis in Manson and in most subsequent cases was on identifying and eliminating such practices, since they were the aspect of eyewitness inaccuracy that could most readily be fixed.

While much remains to be done in this regard, progress has been made. At least nine states now require that line-ups and photo arrays be “blindly” administered, i.e., by a police officer who has no familiarity with the investigation. Many states and localities also require the police managing the line-ups and photo arrays to read from a script that minimizes suggestiveness. Further still, eleven states now require that the eyewitness’s initial degree of confidence in making an identification be recorded and made available to the defense. While further reforms are required—such as videotaping the suspects in the line-ups (and perhaps the eyewitness reactions) and training the police who administer the line-ups and photo arrays to avoid even unconscious suggestiveness—at least police procedure is an area where concrete steps can be taken to minimize inaccuracy.


But neither the courts nor the police have done much to deal with the bigger problem of eyewitness inaccuracy caused by fundamental problems in an ordinary person’s perception and memory. Although the Manson decision invited federal trial courts to exclude eyewitness testimony, not just for police suggestiveness but also for inherent infirmities, this invitation coming as it did before most of the research on the less obvious weaknesses of perception and memory had been undertaken has in practice rarely led to such exclusion.

More recently, however, a few state courts, most notably in New Jersey, have begun experimenting with a different approach: either instructing jurors about the more subtle limitations on human perception and memory that affect eyewitness identification or allowing experts to testify about them. Regretfully, preliminary studies have concluded that the effectiveness of these approaches is modest. Jurors seem to interpret a judge’s special instructions on the subject of eyewitness identification as a veiled message that the judge does not believe the eyewitness; they therefore do not try to distinguish between identifications that are more or less likely to be affected by the factors referenced by the judge, but rather disregard them all. Expert testimony, for its part, often devolves into a battle between experts on both sides, which, according to these studies, the jury resolves by ignoring every one of them.4 In both cases, the result is that jurors still do not undertake the admittedly difficult task of distinguishing reliable from unreliable eyewitness identifications.

In my opinion, these approaches also ignore the fact that the overwhelming majority of criminal cases—more than 95 percent—are resolved by plea bargains or other dispositions not requiring any fact-finding by a judge or jury. Instead, resolving criminal cases has become primarily the responsibility of prosecutors, who now have nearly unfettered discretion to decide who shall be charged, what the charges will be, and how they should be resolved.5 While this is unfortunate, it is also unlikely to change in the foreseeable future. Thus the best thing that can now be done to mitigate the frequent inaccuracy of eyewitness identifications is to educate prosecutors, through training early in their careers, about the dangers of inaccuracy that persist even when police procedures have been unassailable. It may be that courts could not order such training (though the legislative and executive branches easily could); but the courts could suggest its adoption with an emphasis that might be persuasive.

One other modest mitigating factor should be mentioned. For many everyday crimes, like robbery, the presence of surveillance cameras in stores and buildings has made the police somewhat less dependent on eyewitness identification. The broader use of such surveillance cameras should therefore be encouraged. Of course, this is only a partial solution, not only because those committing crimes often seek to avoid detection by, for example, wearing masks, but also because it remains the case that most violent crimes are committed by unmasked individuals in places such as alleys and private homes where videotaping is not occurring.

Eyewitness identification thus presents the legal system with a challenge unlike any other. In many cases, the only direct evidence of who committed a crime is the testimony of an eyewitness. Yet modern science suggests that much of such testimony is inherently suspect—but not in ways that jurors can readily evaluate from their own experience. The result, alas, is a likelihood of wrongful convictions.