The Connection Between Eyewitness Confidence and Accuracy

When a witness takes the stand, looks at the defendant, and says with full conviction, “That’s the person I saw,” it carries enormous weight. Jurors, judges, and prosecutors tend to treat an eyewitness’s confidence as a direct indicator of how accurate their memory is. But decades of psychological research tell a more complicated-and troubling-story. The connection between how sure a witness feels and how correct they actually are is far weaker and more fragile than most people assume.

Why confidence feels like proof 🔗

It’s a natural human instinct to trust someone who appears certain. In courtrooms, eyewitness confidence is often the most influential factor jurors use when deciding whether testimony is reliable. Research consistently shows that jurors are more likely to believe witnesses who appear highly confident, even when those witnesses turn out to be wrong. At the same time, jurors tend to dismiss less confident witnesses who may actually be more accurate.

This reliance on confidence has real consequences. According to the Innocence Project, eyewitness misidentification has contributed to more than 60% of wrongful convictions later overturned by DNA evidence. In many of these cases, the mistaken witness appeared absolutely certain on the stand-even though their identification was completely wrong.

What research actually shows about the confidence-accuracy link 🔗

Early studies examining the statistical relationship between eyewitness confidence and identification accuracy were not encouraging. Researchers measured this relationship using correlation coefficients-a number between 0 and 1 that indicates how closely two things move together. A value near 1 would mean confidence is a strong predictor of accuracy.

What they found was disappointing. Initial research examining eyewitness certainty found correlations generally in the .00 to .20 range, meaning confidence was essentially unrelated to accuracy. These studies bundled all types of witness responses together-whether the witness identified a suspect, chose a filler (an innocent person placed in the lineup), or rejected the lineup entirely.

A meta-analysis by Sporer and colleagues in 1995 examined the existing body of research and confirmed that the overall correlation between confidence and accuracy was modest at best. Similarly, researchers like Bothwell, Deffenbacher, and Brigham found that in some studies the correlation was reasonable, but in others it was small or even nonexistent.

The “pristine conditions” debate 🔗

The conversation shifted significantly with a major 2017 review by John T. Wixted and Gary L. Wells, published in Psychological Science in the Public Interest. They argued that when identification procedures meet specific standards-what they called “pristine” conditions-the confidence-accuracy relationship is actually quite strong.

Pristine conditions include several key requirements. The lineup must be the witness’s first memory test, conducted before any outside contamination. Only one suspect should appear in the lineup, surrounded by known-innocent fillers who match the perpetrator’s description. The lineup administrator should not know which person is the suspect. And the witness’s confidence must be recorded immediately after identification, before any feedback.

Under these tightly controlled circumstances, Wixted and Wells found that high-confidence identifications were surprisingly accurate, and low-confidence identifications were quite error-prone. This finding has been replicated in both laboratory mock-crime studies and police field studies.

Why pristine conditions rarely exist in real life 🔗

Here’s the problem: the conditions Wixted and Wells describe are often not met in actual criminal investigations. A 2024 review by Moore and colleagues at the University of Utah pointed out that while researchers broadly agree on a strong confidence-accuracy link under ideal conditions, such conditions are not consistently present in criminal cases. When real-world factors intrude-unfair lineups, delays between the crime and the identification, cross-racial identification, or exposure to post-event information-the relationship between confidence and accuracy deteriorates significantly.

Furthermore, many jurisdictions have not yet adopted the procedural reforms needed to create pristine testing conditions. Non-blind lineup administration (where the officer conducting the lineup knows who the suspect is) remains common in many parts of the United States, despite strong evidence that it can influence witness choices through subtle, even unintentional, cues.

How confidence gets inflated after the fact 🔗

One of the most troubling findings in this area is that eyewitness confidence is not a fixed quantity. It changes over time-and it is highly susceptible to outside influence.

Post-identification feedback 🔗

The post-identification feedback effect is one of the most well-documented phenomena in eyewitness research. When a witness is told something like “Good, you picked the suspect,” their confidence in the identification increases dramatically-even if their identification was wrong. This was first demonstrated in a landmark study by Wells and Bradfield in 1998 and has been replicated many times since.

Research by Greenspan and Loftus (2020) took this further by showing that even misleading reminders about a witness’s own prior confidence level can shift their retrospective confidence judgments. Participants who were falsely told they had been more confident than they actually were later recalled being more certain at the time of identification. Even participants who received no feedback at all showed significant confidence inflation over a one-week period-a phenomenon the researchers termed “natural confidence inflation.”

A related study found that when mistaken eyewitnesses received confirming feedback, outside evaluators could no longer distinguish between accurate and inaccurate witnesses. Without feedback, evaluators were reasonably good at telling the difference. Confirming feedback essentially erased any visible cues that might have helped jurors detect errors.

Repeated questioning and other influences 🔗

Confidence can also be inflated by repeated questioning, learning that another witness identified the same person, exposure to media coverage, or simply preparing to testify. Each of these factors can increase a witness’s expressed certainty without improving-and sometimes while further degrading-the accuracy of their underlying memory. Critically, research shows that these confidence-boosting effects tend to have their largest impact on inaccurate memories rather than accurate ones.

The courtroom gap: what jurors and judges don’t know 🔗

Despite decades of research, the people who need this knowledge most often lack it. Surveys of U.S. judges found they averaged only 55% correct on a knowledge scale about eyewitness testimony factors. Roughly one-third of surveyed judges incorrectly believed that eyewitness confidence at trial is related to accuracy, one-third correctly recognized it is not, and one-third admitted they did not know.

Jurors fare no better. Studies across multiple countries show that laypersons have limited understanding of eyewitness factors and consistently overvalue confidence while ignoring more diagnostic indicators of accuracy, such as how the identification procedure was conducted. Jurors also struggle to apply whatever knowledge they do possess to the specific facts of a case.

Legal safeguards designed to address this gap-including cross-examination, jury instructions, and even expert testimony-have largely failed to consistently improve juror sensitivity to eyewitness reliability. The most common effect of expert testimony, for instance, is to make jurors more skeptical of all eyewitnesses rather than better at distinguishing accurate witnesses from mistaken ones.

Real-world consequences: when confidence leads to conviction 🔗

The case of Ronald Cotton illustrates how devastating overconfident misidentification can be. Jennifer Thompson was absolutely certain that Cotton had assaulted her. She identified him confidently in a lineup and testified against him at trial with unwavering conviction. Cotton was sentenced to life plus 50 years. It took 11 years and DNA testing to reveal that another man, Bobby Poole, was the actual perpetrator.

Thompson’s misidentification was not the result of carelessness or indifference. She had tried hard to memorize her attacker’s face during the crime. But memory does not work like a video recording. It is reconstructive, selective, and vulnerable to distortion at every stage-during the event itself, during storage, and during each subsequent act of recall. As the Innocence Project notes, each attempt at recall can lead to further alteration of the original memory.

Cases like Cotton’s are not isolated. Nationally, 69% of the 367 DNA exonerations tracked by the Innocence Project involved eyewitness misidentification, making it the single leading cause of wrongful conviction in the United States.

What can be done to close the gap 🔗

Researchers and reform advocates have identified several practical steps that can help align eyewitness confidence more closely with accuracy-or at least prevent confidence from becoming misleading.

Recording initial confidence immediately: If confidence is going to be meaningful, it must be captured at the moment of identification, before any feedback or other contamination. A witness’s confidence expressed at trial-often months or years later-tells us very little about the quality of their original memory.

Using blind administration: The officer conducting the lineup should not know which person is the suspect. This prevents unintentional cues from influencing the witness’s choice or confidence level.

Constructing fair lineups: The suspect should not stand out from the fillers. Everyone in the lineup should reasonably match the description of the perpetrator.

Avoiding confirming feedback: Officers should refrain from telling witnesses whether they selected the “right” person. Even well-meaning comments can inflate confidence and distort memory.

Educating legal professionals: Judges, attorneys, and law enforcement officers need regular, evidence-based training on how memory works and what factors genuinely predict eyewitness accuracy. Several states, including New Jersey, Oregon, and Connecticut, have already reformed their approaches to assessing eyewitness reliability based on modern scientific understanding.

The bottom line 🔗

The relationship between eyewitness confidence and accuracy is not straightforward. Under tightly controlled, ideal conditions, confidence can be a useful indicator-particularly at the highest and lowest ends of the scale. But in the messy reality of criminal investigations, where procedures are often imperfect and memories are vulnerable to contamination, confidence is an unreliable guide. A witness who is 100% sure can still be 100% wrong. Understanding this disconnect is essential for anyone involved in the criminal justice system, from the officers conducting lineups to the jurors deciding a defendant’s fate.

What do you think? How much weight should jurors give to an eyewitness’s confidence when evaluating their testimony? And given what we know about how easily confidence can be inflated, should courts require that initial confidence statements be recorded and presented alongside any later testimony?

References

1. https://pmc.ncbi.nlm.nih.gov/articles/PMC4131297/
2. https://innocenceproject.org/eyewitness-misidentification/
3. https://journals.sagepub.com/doi/full/10.1177/1529100616686966
4. https://www.psychologicalscience.org/publications/journals/eyewitness-identification.html
5. https://psych.utah.edu/_resources/documents/people/committee-docs/fall-2025/2024%20-%20Moore%20et%20al%202024%20Eyewitness%20Confidence%20Does%20Not%20Necessarily%20Indicate%20Accuracy.pdf
6. https://pubmed.ncbi.nlm.nih.gov/32378911/
7. https://pubmed.ncbi.nlm.nih.gov/24341835/
8. https://innocenceproject.org/news/how-eyewitness-misidentification-can-send-innocent-people-to-prison/