# Why is eyewitness testimony unreliable?

> Written with AI · [Psychology: mind, memory and behaviour](https://lessoons.com/en/psychology-mind-memory-and-behaviour) · Grade 11 · 2026-11-11 · https://lessoons.com/en/psychology-mind-memory-and-behaviour/why-is-eyewitness-testimony-unreliable

Eyewitness testimony can be unreliable because memory is rebuilt, not replayed. Each time witnesses recall an event, they fill the gaps from what usually happens (their schemas), from questions asked afterward and from what others said, and cannot tell the parts apart. One word in a question changed speed estimates in Loftus and Palmer's study (1974). Witnesses to real crimes are often accurate, but confidence is not proof.

## The topic explained

### Memory is rebuilt, not replayed

Eyewitness testimony feels like a recording, and that is why it can be unreliable. Reconstructive memory means that recall assembles an event from a fragment that was stored, plus a schema, a mental framework of what usually happens in such a scene, plus anything learned later. The witness experiences the result as one memory. Someone who saw a robbery may "remember" a mask because robbers in films wear one.

Frederic Bartlett showed this in 1932. British participants read The War of the Ghosts, a Native American folk tale, and retold it again and again. The story grew shorter, strange details dropped out, and unfamiliar ones turned familiar: canoes became boats. Bartlett called this an effort after meaning: people bend what they recall toward what makes sense to them. The common misreading is that a distorted memory is a lie; the witness is honestly reporting what the mind rebuilt.

### One word in a question: Loftus and Palmer

In 1974 Elizabeth Loftus and John Palmer showed 45 students seven films of car accidents and asked, "About how fast were the cars going when they hit each other?" Each group heard a different verb. With smashed, the mean estimate was 40.8 miles per hour; with contacted, 31.8. A leading question carries its own answer.

In a second experiment, 150 students saw one crash. A week later they were asked whether they had seen broken glass; there was none in the film. In the smashed group 32% said yes, in the hit group 14%, and in a group with no speed question 12%. The verb had changed the memory, not only the answer. Critics note that the participants were students watching films, with nothing at stake.

### Time, forgetting and memories that never happened

Hermann Ebbinghaus (1885) found that forgetting is fastest at first: relearning a list saved about 58% of the work after 20 minutes but only about a third after a day. Gaps open quickly, and reconstruction fills them with post-event information: what other witnesses said, a news report, the wording of a question. That is why investigators are taught to record a first account early, in the witness's own words.

Elizabeth Loftus and Jacqueline Pickrell (1995) went further: told by a relative that they had once been lost in a shopping mall, about a quarter of 24 participants came to remember it. A preregistered replication with 123 participants (Murphy and colleagues, 2023) found 35%, although other researchers have questioned how a false memory was counted.

### What a court can and cannot conclude

The other side matters too. John Yuille and Judith Cutshall (1986) re-interviewed 13 witnesses to a real shooting in Vancouver four to five months later: their accounts were detailed and accurate, and leading questions did not mislead them. Real events, with real stakes, may be remembered better than films in a laboratory.

Still, eyewitness misidentification played a part in 69% of the first 375 DNA exonerations in the United States (Innocence Project, 1989 to 2020). Confidence is not proof: when told "Good, you identified the suspect," witnesses who had all picked the wrong person grew more confident (Wells and Bradfield, 1998). Research like this shaped the cognitive interview, developed by Ronald Fisher and Edward Geiselman from 1984, which replaces leading questions with open ones. A good answer weighs both kinds of evidence instead of calling all witnesses unreliable.

## Practice: reconstructive memory and eyewitness testimony

For 11th grade psychology students. Answer each question from what is printed on the sheet, then check your work against the answers.

1. **In your own words, what is reconstructive memory, and what is a schema?** — Reconstructive memory: recall rebuilds an event from a stored fragment, a schema and later information, instead of replaying a recording. A schema is a mental framework of what usually happens in a kind of situation.
2. **A student retells The War of the Ghosts and writes boats instead of canoes, and leaves out the strange parts. Explain the change with Bartlett's idea.** — The student's schema of a familiar story replaced the unfamiliar detail with a familiar one and dropped what did not fit. Recall was rebuilt to fit what the student expects, as Bartlett found in 1932.
3. **Read the bar chart. Which verb produced the highest estimate and which the lowest? By how many miles per hour do they differ?** — Smashed gave the highest mean (40.8 mph) and contacted the lowest (31.8 mph). The difference is 40.8 − 31.8 = 9.0 mph, from one word in the question.
4. **In the second experiment there was no broken glass in the film. Work out the missing percentage for the smashed group. What does the table show?** — 16 of 50 is 32%. More than twice as many in the smashed group as in the hit group (14%) remembered glass that was never there, so the verb changed the memory itself, not only the speed estimate.
5. **Read Ebbinghaus's curve: about how much was saved after 1 day? When is forgetting fastest, and why does that matter for a witness questioned weeks later?** — About 34% (33.7%) after one day. Forgetting is fastest in the first hours and then slows. A witness questioned weeks later has many gaps, which schemas and later information can fill.
6. **In 1995 about 25% of 24 participants came to remember a childhood event that never happened. In 2023 a preregistered replication with 123 participants found 35%. Why does the replication matter?** — A finding from 24 people could be chance. A planned replication with five times as many participants, with its analysis registered in advance, found a similar effect, which makes it more trustworthy, although researchers still debate how a false memory should be counted.
7. **Yuille and Cutshall (1986) found that witnesses of a real shooting were accurate months later and resisted leading questions. Give one reason their results differ from Loftus and Palmer's.** — A good answer names a real difference: a real, emotional event with real consequences versus films in a laboratory (ecological validity); or the student participants in Loftus and Palmer. Yuille and Cutshall's study was also small (13 witnesses) and could not control conditions.
8. **A witness says in court, "I am completely sure that is the man." What may the court conclude, and what not?** — A good answer says that confidence alone does not prove accuracy (Wells and Bradfield, 1998, showed it can grow after feedback), asks how and when the identification was first made and what the witness was told, and weighs the identification with other evidence, without dismissing every witness.

## Common questions

### Is eyewitness testimony always unreliable?

No. Studies of real crimes, such as Yuille and Cutshall's (1986), found witnesses who were detailed and accurate months later. Testimony is most at risk from leading questions, long delays, talk with other witnesses and feedback after an identification.

### What is a leading question?

A question whose wording suggests the answer, such as using smashed instead of hit, or asking about the broken headlight when nobody mentioned one. Police guidelines in many countries now favor open questions for this reason.

### What did Loftus and Palmer find?

In 1974 they showed that one verb in a question changed speed estimates for the same films, from 31.8 miles per hour with contacted to 40.8 with smashed, and that a week later more of the smashed group remembered broken glass that was not there.

### Can a false memory be planted?

In studies, some people come to remember invented events, such as being lost in a mall as a child: about 25% in 1995 and 35% in a 2023 replication. Researchers still debate how complete these memories are and how far the findings reach.

### What is the forgetting curve?

Hermann Ebbinghaus's 1885 measure of forgetting, made on himself with lists of nonsense syllables. Most forgetting happens soon after learning, then it levels off: he saved about 58% of the work after 20 minutes, a third after a day and a fifth after a month.
