Peak–end rule

The peak–end rule describes a conditional pattern in remembered experience. A person's later summary can lean more heavily on an episode's most intense point and closing stretch than on the rest. That judgment may differ from one that weights every moment by its duration. The rule is a useful predictor in some settings, not a law of memory and not proof that the rest of the episode disappeared.

Researchers sometimes average the peak and end ratings to predict a person's global evaluation. That number is an analytical model. The evidence does not show that every mind literally computes (peak + end) / 2.

Illustrative hypothetical: Two rehearsals under the dome

A volunteer at a community planetarium is trying to calibrate a stubborn dome projector. They run two 18-minute rehearsals and record their frustration every two minutes.

Both records produce an identical mean frustration score and share the same maximum. One rehearsal ends just as the star field locks into place. The other ends while the image is still drifting.

The next morning, the volunteer rates each rehearsal as a whole and says which setup they would rather repeat. If the rehearsal with the improving ending receives the better rating, that result would illustrate end weighting. It would not prove that the average, duration, or earlier moments never matter.

This is a constructed example, not a study result. Real experiments rarely hold the mean, peak, duration, and sequence this neatly apart. Changing the finish can also turn a flat or worsening profile into an improving one, and sequence itself affects judgments of extended profiles (Varey & Kahneman, 1992).

What counts as the peak and the end?

The peak is the most extreme value on the scale used in a study. For an aversive episode, it might be the moment of worst pain. For a pleasant episode, it might be the moment of greatest pleasure. The end may be the final rating or an average over a defined final segment.

Early studies of extended profiles, films, and discomfort found that a few moments could predict a later summary better than a duration-weighted total in those tasks. Fredrickson and Kahneman's film studies, for example, compared momentary affect with later evaluations of pleasant and aversive clips. Duration had only a small influence under some retrospective conditions, but the authors did not claim that peaks and endings exhaust every determinant of remembered affect.

The safest way to read the rule is descriptive: peak and end values sometimes predict a retrospective judgment. These studies do not establish one universal process in which emotional events are stored, made accessible, and retrieved according to those two moments.

Experienced utility is not remembered utility

A momentary record answers, “How did this feel as it happened?” A later global rating answers, “How do I judge the episode now?” A repeat choice asks a third question: “Which option will I choose again?” Kahneman, Wakker, and Sarin called the first kind of measure instant or experienced utility and the later summary remembered utility.

Those measures can disagree without either one being fake. The difference matters because a pleasant memory does not erase difficult minutes that were actually experienced. A repeat choice also reflects more than memory, so it should not be treated as a perfect readout of what a person remembers.

What the best-known experiments found

In a small cold-water study, 32 male university students experienced two painful episodes. One lasted 60 seconds. The other added 30 seconds during which the water remained painful but became slightly warmer. Twenty-two participants chose to repeat the longer episode. In a separate replication without continuous ratings, 24 of 37 chose it. They correctly identified the longer exposure, which rules out a simple failure to know which episode lasted longer (Kahneman et al., 1993).

One randomized trial extended this question to a specific medical setting. Researchers assigned 682 colonoscopy outpatients either to the usual procedure or to one with a brief added closing phase that was generally gentler than the preceding minutes. The added phase lengthened the procedure, yet the intervention group later rated it as less unpleasant and had a small adjusted increase in return over follow-up (Redelmeier, Katz, and Kahneman, 2003).

That trial supports a bounded claim about one procedure. It did not establish greater accumulated pain for each patient, a rule for all medical experiences, or a sound general strategy of prolonging discomfort. Experienced welfare across the full procedure remains ethically relevant.

Does the peak–end rule always work?

No. The pattern is substantial across parts of the literature, but it is heterogeneous. A 2022 meta-analysis assembled 174 effect sizes and was followed by a 2024 corrigendum that changed the sign-coding rule. Read together, the papers indicate that peak-end summaries can predict retrospective evaluation, while the overall average can be comparably predictive (Alaybek et al., 2022; 2024 corrigendum). The original article's superseded headline estimate should not be used.

Evidence for positive episodes is especially mixed. Pleasant-film studies found snapshot weighting under controlled conditions. Across three candy-and-toy experiments enrolling 988 people from childhood through adulthood, researchers detected no peak-end pattern (Mah and Bernstein, 2019). Studies of actual and imagined meals found duration neglect but minimal peak, end, or primacy effects (Rode, Rozin, and Durlach, 2007). In reconstructed multiepisode days from U.S., French, and Danish samples, the duration-weighted average was the strongest predictor, and the ending did not predict the global evaluation (Miron-Shatz, 2009).

So a pleasant peak and a pleasant ending do not guarantee a better overall memory regardless of length or average pleasure. The result depends on the episode, the population, how the moments are sampled, when the summary is collected, and what the participant is asked to report.

Peak–end weighting and duration neglect are separate

Duration neglect means that a later judgment changes little when an episode lasts longer. Peak–end weighting means that the extreme and final moments receive disproportionate weight. Either pattern can appear without the other. The meal studies are a clear example: duration mattered little, yet peak and end effects were also minimal.

Duration sensitivity can return when the comparison or rating format makes length more relevant (Ariely and Loewenstein, 2000). Duration can also matter indirectly because the same peak or ending occupies a different temporal position in a longer episode (Weingarten, Zauberman, and Diehl, 2021). These findings are hard to reconcile with a simple claim that memory fails to preserve time.

The outcome matters too. Immediate ratings, delayed recollections, rankings, and repeat choices are different measures. In one 28-day rheumatology study, peak and end values added a small amount of prediction beyond the daily mean for recalled pain but not fatigue (Schneider et al., 2011). Results from North American students or selected clinical patients do not establish a culturally universal tendency.