Eyewitness memory is reliable - people accurately remember what they see, especially under stress.

For most of the twentieth century, eyewitness testimony held special authority in court. A witness who had seen the crime, recognized the face, remembered the weapon, and testified with confidence seemed to offer a direct record of reality. Memory was often treated as a recording: imperfect perhaps, but fundamentally stable once formed.

That assumption began to collapse in the 1970s. In 1974, Elizabeth Loftus and John Palmer published their car crash experiment, showing that a single verb in a question could change what people reported seeing. Participants who were asked how fast the cars were going when they had "smashed" into each other gave higher speed estimates than those asked when the cars had "hit" each other. In a follow-up experiment, people exposed to the word "smashed" were also more likely to report broken glass that had never appeared in the original footage. The finding was unsettling because the manipulation was so slight: post-event information, introduced through ordinary questioning, could alter memory itself.

Loftus and later researchers showed that this was not an isolated effect. Eyewitness memory is reconstructive rather than replayed like a video. What a witness recalls can be shaped by leading questions, discussions with other witnesses, news coverage, stress during the event, and the way police administer lineups. High stress often reduces identification accuracy even when witnesses feel intensely certain. The presence of a weapon draws attention away from the perpetrator's face. Cross-racial identifications are, on average, less accurate. Most importantly, a witness's confidence does not reliably track whether the memory is correct.

One of the most famous examples came a decade later. In 1984, Jennifer Thompson was raped in her apartment in Burlington, North Carolina. Determined to survive and identify her attacker, she deliberately studied his face. The next day she chose Ronald Cotton from a photo lineup and later identified him again in person. At trial she testified with complete certainty, and Cotton was convicted. He spent eleven years in prison before DNA evidence exonerated him in 1995 and identified the real perpetrator, Bobby Poole. Thompson had not been lying or guessing. She had been sincere, detailed, confident, and wrong.

Cases like Cotton's helped expose the gap between psychological science and courtroom practice. Research on eyewitness unreliability had been accumulating since the early 1970s, but the legal system was slow to absorb it. In 1979, the U.S. Supreme Court ruled in Manson v. Brathwaite that juries could assess eyewitness reliability for themselves rather than requiring stricter judicial standards. But the science was showing that unreliable and reliable memories often feel exactly the same from the witness's point of view.

By the 1980s and 1990s, the evidence was overwhelming enough that reform recommendations began to emerge. A 1999 National Institute of Justice guide urged police to use procedural safeguards such as double-blind lineup administration, sequential presentation of suspects, and standardized instructions making clear that the perpetrator might not be present. Even then, reform was uneven. DNA exonerations later reinforced the point with brutal clarity: eyewitness misidentification proved to be one of the leading contributors to wrongful convictions.

The collapse of faith in eyewitness certainty was not a discovery that people are dishonest. It was a discovery about the architecture of memory. A person can study a face, try to remember carefully, testify sincerely, and still be mistaken. What changed after 1974 was not human nature, but scientific understanding: memory is not a fixed recording of the past. It is a reconstruction, and reconstructions can be wrong.