Case File 262 · Investigation Psychology

How Reliable Is Eyewitness Testimony?

By Murder Book Editorial TeamPublished by Gungrounds, the studio making Murder BookPublished 3 October 2026Updated 3 October 2026

Eyewitness testimony is useful but far less reliable than it feels, to the witness or to the people hearing it. Memory is not a recording. It is reconstructed each time it is recalled, and it can be incomplete at the moment of the event, distorted while it is stored, and altered during questioning. Research over more than forty years has shown that honest, confident witnesses are sometimes wrong, and that their errors have contributed to wrongful convictions. None of this means eyewitnesses should be ignored. It means their accounts should be gathered carefully, treated as evidence to be tested rather than accepted whole, and checked against independent facts. The reliability of any particular account depends on the conditions of the event, the time since it happened, and how the witness was questioned.

Key takeaways

Eyewitness memory is reconstructive and can change between the event and recall.
Eyewitness misidentification has been a leading contributing factor in DNA exoneration cases.
Confidence does not guarantee accuracy; a sure witness can still be mistaken.
Reliability depends on viewing conditions, delay, stress and how questions are asked.
Accounts are most trustworthy when taken early, with open questions, and corroborated.

Why is eyewitness testimony treated with caution?

The strongest reason comes from cases where the facts were later settled by other evidence. Analysis by the Innocence Project found that eyewitness misidentification was a contributing factor in roughly 69% of the DNA exoneration cases it reviewed in the United States, the most common factor among them. These were not cases of lying witnesses. They were people who believed they were identifying the right person and were wrong.

In 2014, the US National Research Council reviewed the science in a report titled Identifying the Culprit: Assessing Eyewitness Identification. It summarised what research does and does not establish about eyewitness identification and recommended changes to how such evidence is collected and presented, including better line-up procedures and clearer guidance for courts. The report reflects a broad scientific consensus that eyewitness evidence is valuable but fallible, and that its handling affects its accuracy.

What makes a memory more or less reliable?

Gary Wells and Elizabeth Olson, in a 2003 review in the Annual Review of Psychology, distinguished two kinds of factors. System variables are things the justice system controls, such as how a line-up is run or how a witness is questioned. Estimator variables are things it cannot control, such as lighting, distance, duration and stress at the time of the event.

Several conditions are known to weaken accuracy:

Poor viewing conditions: brief glimpses, darkness, distance or obstruction.
Delay: memory fades over time, discussed in How Does Time Affect Eyewitness Memory?.
Stress: high stress at encoding is associated with worse recall and identification, covered in Does Trauma Make Memory More or Less Accurate?.
Post-event influence: leading questions, media and other witnesses can reshape an account, as set out in Can New Information Change an Old Memory?.

Does a confident witness mean an accurate one?

Not on its own. Confidence feels like a signal of accuracy, which is why juries find sure witnesses persuasive, but the relationship between the two is weaker and more fragile than intuition suggests. Confidence can also grow after the event, for reasons unconnected to the quality of the original memory, such as feedback or repeated retelling. The gap is explored in Does Eyewitness Confidence Mean the Memory Is Accurate?.

Research on line-ups suggests that confidence may be more informative at the very first identification, under careful, unbiased conditions, than later in the process when it has had chances to inflate. That is one reason the National Research Council stressed recording a witness's confidence at the moment of identification.

How can accounts be gathered more reliably?

Because many of the biggest risks are system variables, careful procedure makes a real difference:

Interview as soon as possible, before the witness is exposed to other sources.
Use open, non-leading questions and let the witness recall freely before asking for specifics.
Keep witnesses apart, so accounts do not contaminate one another.
Run fair line-ups, with proper instructions and fillers, and record confidence straight away.
Record the interview, so the origin of each detail can be traced.
Seek corroboration, treating the account as one strand of evidence among several.

These steps do not make memory infallible. They reduce the avoidable errors so that what remains is a cleaner reflection of what the witness actually experienced.

What about identification line-ups specifically?

Picking a face from a line-up is one of the highest-stakes things an eyewitness does, and it is especially sensitive to how the procedure is run. The National Research Council's 2014 review highlighted several procedural safeguards that affect accuracy: giving witnesses a clear instruction that the culprit may not be present, using fillers who fit the description so the suspect does not stand out, and recording the witness's confidence at the moment of the identification rather than later.

The review also pointed to the value of administrators who do not know which person is the suspect, so they cannot, even unintentionally, steer the witness. These are system variables in Wells and Olson's terms, which means they are within the justice system's control. Improving them does not guarantee a correct identification, but it removes avoidable sources of error and keeps post-identification feedback from inflating a witness's confidence after the fact.

Is eyewitness testimony worth anything, then?

Yes. The point of the research is not that witnesses are useless but that their accounts are evidence, with strengths and weaknesses like any other. A well-gathered account from a witness who had a clear, unhurried view, questioned early and openly, can be valuable. A fragmentary account taken late, after the witness has read coverage and compared notes with others, deserves more caution. Judging reliability means looking at the conditions, not just the certainty of the person speaking.

Crucially, inaccuracy is usually honest. A mistaken witness is typically not lying; their memory has failed in ordinary, well-documented ways. So an error in one detail does not prove dishonesty, and confidence in another does not prove truth. Corroboration settles what memory alone cannot.

Is there an upcoming game about weighing unreliable accounts?

Murder Book: The Art of Interrogation makes the fragility of testimony its central subject. It is an upcoming narrative interrogation game from Gungrounds, an independent studio in Zagreb, Croatia, inspired by real police footage and psychological interviews. Instead of handing players a reliable narrator, it presents unreliable stories, narrative gaps and contradictions to work through.

That mirrors the research: a confident account is a starting point, not a conclusion. The game rewards reading how a story holds up, comparing statements against each other and against the evidence, and noticing where certainty outruns support. Cases are designed for focused 5–10-minute sessions. Murder Book is in development for PC (Steam) and mobile, with release currently planned for the end of Q2 2027.