Human memory is, at its core, an unreliable witness. More than 70 percent of all convictions of people who were exonerated with DNA evidence between 1989 and 2019 were cases of mistaken eyewitness identification. Meanwhile decades of research shows that memories can easily be warped each time they’re recalled—a truth that suggests eyewitness accounts should be taken with enormous grains of salt. But it might not be so simple: new studies are bringing nuance to whether and how eyewitness testimony should be used.
Memory is malleable, and even worse, humans are terrible at knowing whether their memories are reliable. Nevertheless, memory can be reliable evidence: It is “like any other piece of forensic evidence,” says Laura Mickes, a memory researcher at the University of Bristol in England. “If it’s contaminated, it’s not going to be good.”
Malleable Memory
On supporting science journalism
If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.
The problem with memory is that recollecting an event isn’t like taking a book off the shelf to read and then putting it back just as you found it. Instead, each time we bring a memory out of long-term storage, it goes through a process called reconsolidation, which allows new information to seep in and alter the memory.
In the early 1970s memory researcher Elizabeth Loftus of the University of California, Irvine, demonstrated that simply by asking the same question in different ways, she could get witnesses to a car crash to remember the event in distinct ways. Using the word “smashed” in a question versus “collided” or “bumped,” for example, was more likely to prompt witnesses to remember seeing broken glass that hadn’t actually been there. Other research on “flashbulb” memories, or memories of major events, has also shown that although people may believe they will never forget exactly where they were and what they were doing on September 11, 2001, they get about 40 percent of the details wrong.
For eyewitness accounts, factors such as how far a witness was from the crime, whether the scene was well-lit and even whether the witness and perpetrator were the same race can affect how accurate a witness’s memory might be.
Some of these factors are less well understood than others, making them harder to account for. For example, research into how intoxication affects memory is fairly new, says Lilian Kloft-Heller, a forensic psychologist at Maastricht University in the Netherlands. Some evidence indicates that people who witness a crime while drunk or high report fewer details but can still be accurate in the details they report, Kloft-Heller say. “They can still give valid testimony,” she says. “What is most important is that they are questioned well.”
While there are still many open questions about the best way to question a witness, memory researchers have developed evidence-based recommendations for police to ensure they don’t inadvertently contaminate eyewitnesses.
Collecting the Evidence
One of the easiest ways to mess up a witness’s memory is to subject them to a poorly done police lineup, says John Wixted, a memory researcher at the University of California, San Diego. Imagine witnessing a mugging. After the crime, the police bring you a man in handcuffs—a single suspect presented to you in what criminal justice experts call a “showup.” You might not be sure he’s the guy, but you tell the cops that he might be. A day or a week later, they call you in and present a photo array of six suspects, including the man at the showup. His face pings your memory—and your memory places him at the crime scene. Yet you’re really remembering him from the first time the police showed him to you, not from the mugging itself.
“The memory has changed, and the witness doesn’t know it,” Wixted says.
Memory contamination can lead witnesses to confidently identify a perpetrator in the courtroom even if they were far less certain immediately after the crime. Wixted and his colleagues have found that confidence in court is a poor indicator of a witness’s accuracy. But if witnesses are shown only a fair photo lineup after a crime, then the degree of confidence they have in their memory tends to indicate accuracy.
In fact, Mickes and Wixted have found that, in many exoneration cases, a witness resolutely identified a person in court despite having dithered and debated during the initial photo lineup. Researchers now recommend that police not only ask a witness to identify a perpetrator but also ask how confident they are in that ID.
“When they pick someone out of the lineup with a high confidence, they are more likely to be accurate than people who pick someone with low confidence,” Mickes says. “That’s really meaningful, and it’s something we should be listening to.”
Eyewitness memory researchers are turning toward other clues that might indicate accuracy. For example, work by Andrew Smith, a psychologist at Iowa State University, suggests that combining multiple behaviors, such as how quickly the witness picks out the suspect, whether they use process of elimination and how much they debate with themselves, can offer a clearer picture of whether they’re reliable.
Smith is also studying the information police glean from witnesses when they fail to identify anyone in a lineup. Often, he says, asking witnesses about how assured they are in these cases can yield useful information, including whether the witness is extremely confident that the police’s suspect is not the culprit. “You get these spontaneous pieces of information that makes it clear this witness knows exactly what she’s talking about,” Smith says.
How to Conduct a Fair Lineup
The story of Ervin Harris illustrates the importance of perfecting lineup procedures. In 2026 Harris was exonerated after spending 42 years in prison for being convicted of a 1974 rape. The victim described a perpetrator who was 24 to 25 years old, skinny and clean-shaven, with a clear speaking voice. Police put together a photo lineup that included at least one man with facial hair, breaking with the best practice that everyone in a lineup meet the witness’s description. The victim chose 19-year-old Harris by a process of elimination, after which police assured her that they’d “go get him”—a statement that could make an initially uncertain witness feel more confident by the time of trial.
The police broke another key rule of suspect identification: They then put Harris in a live lineup, meaning he was the only person in the group whom the victim had seen before. Several other suspects in this lineup also failed to match the victim’s initial description. Harris was not clean-shaven and spoke with a strong stutter, yet the victim identified him as the perpetrator. His conviction at trial was based almost entirely on the victim’s ID; there was no forensic or other evidence tying him to the crime.
“We shouldn’t blame eyewitnesses,” says Travis Seale-Carlisle, a lecturer and memory researcher at the University of Aberdeen in Scotland. “If anything, we should blame the procedures, how the procedures were conducted. That is what has given eyewitness memory a bad name.”
In 2020 experts in the field agreed on the nine evidence-based procedures for conducting a fair lineup, including choosing people who match the suspect’s basic description, only asking a witness to look at a lineup once and asking witnesses how confident they are in their decisions immediately after. The process should also be double-blind, with lineups administered by officers who don’t know who the suspect is.
Yet given that police departments are locally run and follow different rules, how these procedures are applied can be a patchwork. In the U.S., federal guidelines, updated in 2017, recommend police collect information on witness confidence, while in the U.K., that isn’t the case, Seale-Carlisle says. And translating new scientific findings into practice can be slow, he says, especially when researchers are working through contradictions in the evidence. But as memory experts are gaining greater understanding of how our recollection works, he says, “this understanding is trickling into policy.”