The Peak-End Rule: Why the Worst Moment Defines the Memory

That result is genuinely strange. More total discomfort, experienced as less. It is the clearest evidence that remembering and experiencing are separate processes that can disagree.

The two selves disagree often enough that the disagreement has practical consequences. Asked how a holiday was, you consult the remembering self. Asked whether you are enjoying yourself right now, you consult the experiencing one. The two answers can differ sharply for the same fortnight, and neither is lying.

Two selves

Daniel Kahneman frames this as a distinction between the experiencing self, which lives through moments as they occur, and the remembering self, which decides afterwards what the episode was like.

The remembering self does not average the moments. It takes a sample — roughly the most intense point and the final moments — and treats that as the episode.

Everything else, including how long the experience lasted, contributes far less than it should. This is duration neglect, and it is the counterintuitive half of the finding.

The cold water study

Kahneman, Barbara Fredrickson, Charles Schreiber and Donald Redelmeier ran the cleanest demonstration in 1993.

Participants immersed a hand in painfully cold water for 60 seconds. In a second trial they did the same for 60 seconds, followed immediately by a further 30 seconds during which the water was warmed very slightly — still unpleasant, but less so.

The long trial contained strictly more discomfort. Every second of the short trial occurred in it, plus thirty additional seconds of pain.

Asked which they would repeat, a clear majority chose the long trial.

They were not confused about the facts. The ending was better, and the ending dominated the memory.

One caution about how this gets reported. The colonoscopy work is sometimes summarised as showing that doctors should deliberately prolong unpleasant procedures, which overstates it considerably. The extension tested was brief, involved no additional instrument movement, and was evaluated in a randomised trial — a specific intervention, not a general licence.

The clinical version

Redelmeier and Kahneman then tested it where it mattered, with patients undergoing colonoscopy under the less comfortable conditions of the early 1990s. Patients reported discomfort in real time throughout the procedure.

Retrospective evaluations tracked the peak and the final moments, and were largely insensitive to how long the procedure had lasted — which ranged from a few minutes to over an hour.

A follow-up trial extended some procedures by leaving the instrument in place, stationary, for a short additional period at the end. This is uncomfortable but less so than the procedure itself, and it improved patients' retrospective evaluations.

The reason this is more than a curiosity is what it predicts about behaviour: people decide whether to return for future screening based on the remembered version, not the experienced one.

A further boundary concerns what counts as the end. In short laboratory episodes the final moments are unambiguous, but a holiday, a course or a relationship has no single ending — which is part of why extending the rule to long experiences is harder than it looks, and why summaries that apply it to whole life periods are reaching well beyond the evidence.

How firm is it?

The core finding has held up reasonably, and the caveats are about scope rather than existence.

Duration neglect is not absolute. When people are prompted to attend to duration, or when the difference in length is very large, duration does influence evaluations. The original studies were designed so that length varied without being salient.

The rule also describes an average pattern rather than a formula. Peak and end account for much of the variance in remembered affect, not all of it, and the balance between them varies across studies and types of experience.

And there is a boundary worth naming: most demonstrations use relatively short, single-episode experiences with a clear beginning and end. How far it extends to long, complex or repeated experiences is less well established than popular summaries imply.

Why memory takes a sample rather than a sum

The obvious question is why memory would work this way, given that it produces demonstrably wrong answers about which experience was worse.

The usual account is economy. Storing a moment-by-moment record of every episode would be enormously expensive, and most of what memory is for is guiding future decisions — approach this again, or avoid it. A representative sample answers that question cheaply.

Duration may also be genuinely less informative than intensity for that purpose. How bad something got, and how it resolved, arguably say more about whether to repeat it than how many minutes it occupied.

That is a reasonable functional story rather than a demonstrated one, and it should be read as such. What is established is the pattern, not the reason for it.

What it means in practice

The uncomfortable implication is that optimising an experience and optimising the memory of it are different projects, and they can conflict.

A holiday with one extraordinary day and a good ending may be remembered more fondly than a uniformly pleasant one that fizzles out — even though the second contained more enjoyment moment to moment. Which one was better depends on which self you are asking.

The applied version is straightforward and slightly cynical: endings carry disproportionate weight, so the last five minutes of a meeting, a meal, a course or a visit matter more than their duration would suggest.

It also explains something this site has already touched. A frightening experience with a strong peak and a clean resolution is remembered as enjoyable, which is part of why people pay to be scared. And it belongs with the broader finding that remembered time is constructed rather than measured — see why the last decade feels shorter than it was.