DNA testing has been something of a mixed blessing for prosecutors. Provided the samples are handled correctly, a positive test on hair, blood, semen, or other biological evidence can conclusively put a suspect at a crime scene. But the scientific certainty of DNA testing has also exposed just how flawed other arrows in the prosecutor’s quiver really are. DNA exonerations have called into question the accuracy of ballistics analysis, bite mark evidence, hair and carpet fiber evidence, shoe print analysis, jailhouse informants, and even fingerprint identification, once the gold standard of the forensics world.

The latest form of evidence to come under scrutiny is eyewitness testimony. Psychologists have long known about the fallibility of human memory. As far back as 1971, England’s Criminal Law Review Committee warned that over-reliance on eyewitness testimony could lead to false convictions. Going back even to the 1800s, famed psychologist Hermann Ebbinghaus’s memory research established the “Forgetting Curve,” which plots how human recollection fades over time, beginning within minutes of the creation of a memory.

Nevertheless, eyewitness testimony remains a vital part of the criminal justice system, and with good reason. It’s the most abundant form of evidence, and it would be nearly impossible to convict guilty people without it. The problem is that it has for far too long been used irresponsibly, without instituting proper controls to ensure that eyewitnesses aren’t prodded into false recollections, that jurors aren’t permitted to give eyewitnesses more weight than good science allows, and that jurors are made aware of the limits and fallibility of human memory.

DNA testing has thrust the shortcomings of eyewitness recollection back into the spotlight. The cases of Timothy Cole and Ronald Cotton in particular have renewed the discussion. Cole was convicted in 1986 of a rape he didn’t commit. Though the real perpetrator confessed to the crime in 1995 and maintained his confession for years after, Cole’s name wasn’t officially cleared until yesterday. Unfortunately, Cole died in prison in 1999. His family says he couldn’t get proper treatment for his asthma while incarcerated.

Cotton was convicted of raping college student Jennifer Thompson in 1984, due entirely to Thompson’s identification of him in a police lineup. Cotton was also eventually exonerated by DNA testing and—in a rare happy ending for one of these cases—he and Thompson have since reconciled and now advocate together for criminal justice reform. They've just written a book together, and were the subject of recent reports by 60 Minutes and Newsweek.

According to a study published earlier this year in the journal Law and Human Behavior, false eyewitness testimony contributed to 77 percent of the 230 wrongful convictions exposed by DNA evidence over the last decade (the number of exonerations has grown since the study was conducted). These of course are only those cases for which DNA testing was available, which are usually murder and rape cases—crimes for which, generally speaking, there is also usually other evidence available. In crimes where investigators are more likely to rely only on eyewitnesses, robberies or muggings, for example, it’s likely that the problem is even more pronounced.

Psychologists and criminologists have known about these problems with eyewitness testimony long before advances in DNA testing proved them. But it's even more troubling to consider that eyewitnesses tend to become more confident in their identifications with positive feedback.

In a 1999 study, Iowa State researchers Gary Wells and Amy Bradford showed participants grainy video footage of a real case in which a man shot and killed a security guard while robbing a convenience store. They were then given a spread of five pictures, and told that the culprit was included in the photo set. Every one of the participants claimed they could positively identify the culprit. They were all wrong. The researchers had deliberately excluded his photo from the lineup. More troubling still, when one group of participants was given positive feedback from the researchers, that group became more confident in their identifications. Half said they were now “certain” of their identification. Those participants also said they would be more willing to testify against the suspect. They were more likely to describe the security footage as “clear” than other participants and, notably, also denied that the positive feedback had any effect on their identification.

A study released this year takes Wells and Bradford’s experiment even further. Psychologists Lisa Hasel of Iowa State University and Saul Kassin of John Jay College staged a laptop theft in front of a group of students. The students were then shown a lineup of possible suspects. The lineup did not include the actual thief. The students weren’t told they had to pick someone, only to pick the suspect if they recognized him. They were then asked to rate their confidence in their selection from one to 10. Just 33 students correctly said that none of the photos was a match; 173 identified a suspect from the lineup.

The researchers then brought the students back two days later. Some of the students were told that one of the suspects had confessed. Half the students who originally (and correctly) refused to finger a suspect from the lineup changed their minds, now asserting that the person who confessed was indeed the person they saw. Of those who identified the suspect who later confessed, their confidence level in their identification increased from a six to an 8.5. It’s important to note that the students weren’t asked to rate their confidence in the suspect’s guilt, only in their ability to identify him from memory. Even though memory fades over time, the false confession made them more confident in their recollection.

All of this research should tell us that we need to institute reforms. Even subtle, unintentional feedback from police or prosecutors can lead to false identifications. Witness and photo lineups should be double-blind, where neither the officer conducting the lineup nor the witness knows which person is the suspect. Lineups should also include people that the police know are innocent. If a witness selects a known innocent, police and prosecutors will then know that particular witness’s memory isn’t reliable enough to be used as evidence.

Unfortunately, neither the spate of DNA exonerations nor the research on eyewitness identification has changed how police investigate crimes or how prosecutors try them. Though a few major police departments, most notably in Dallas, are considering some reforms, Stephen Saloom, policy director for the Innocence Project, told Nature last year that despite all of the problems with eyewitness identification, when it comes to lineups, "The majority of jurisdictions are simply sticking with what they have always done."

After a string of high-profile criminal justice scandals, the state of California assembled a blue ribbon panel of former judges and prosecutors, criminologists, and defense attorneys to recommend reforms to guard against wrongful convictions. One of the key suggestions was to change the way the state’s police agencies conduct lineups. The reform proposals were twice passed by the state legislature, but after heavy lobbying from the state’s district attorneys and police organizations, they were twice vetoed by Gov. Arnold Schwarzenegger.

Reform prospects took another hit in 2006 when an Illinois study of 700 eyewitness identifications claimed that double-blind, sequential lineups (the type recommended by reformers) produced more errors than lineups where a police officer conducting the lineup knows which person is guilty. The problem is that the study was conducted by Sheri Mecklenburg, general counsel for the superintendent of the Chicago Police Department, an agency long opposed to changing old police procedures.

Mecklenburg’s report was widely derided by psychologists and criminologists for its lack of academic rigor and biased methodology. The critics’ complaints are too numerous too recount here, but the Mecklenburg report's most egregious error was that it calculated a witness’s selection of the police suspect as a “correct” identification. Thus the report counted every Illinois DNA exoneration as a “correct” identification. That’s a considerable oversight, given that the reason the Illinois legislature commissioned the report in the first place was as a response to the state’s high-profile string of wrongful convictions.

The Mecklenburg Report’s main effect was to slow the growing momentum for reforming the way eyewitness testimony is solicited and used in courtrooms. Though it has since been largely discredited, the damage was done.

Perhaps the renewed media interest in eyewitness testimony will prompt lawmakers around the country to revisit the issue. There’s no question that witnesses are an important part of the criminal justice system. But there’s also no debating that when used improperly—as it often is—eyewitness testimony can do an incredible amount of damage.

Radley Balko is a senior editor for Reason.