Skip to content

Memory and forgetting

The forgetting curve

How learning works · Published · Evidence: Strong · About this guide

The forgetting curve describes how new learning fades: quickly at first, then more and more slowly. Ebbinghaus measured it on himself, in a study published in 1885, as the work saved when relearning lists of nonsense syllables, not as the share he could still recall, and a 2015 replication found a similar shape [1][2]. Each study had one participant, so the numbers do not predict what you will forget.

Most of us have read a chapter, felt sure we knew it, and found a week later that much of it had slipped away. That experience has a shape, and it is one of the oldest findings in psychology: forgetting is fast at first and then slows down. This article looks at where that curve comes from, what it does and does not measure, and why it levels off.

Where does the forgetting curve come from?

Around 1880 the German psychologist Hermann Ebbinghaus ran a long series of memory experiments with a single participant: himself [1]. He learned lists of short, made-up syllables, chosen so that what he already knew would help as little as possible. He studied each list until he could recite it twice in a row without a mistake, waited a set time (from about 20 minutes to 31 days) and then learned the same list again to the same standard.

The second learning was what he measured. If the list was still partly in memory, it should take less work to get it back to that standard, and the difference tells you how much of the first effort was left.

What did Ebbinghaus actually measure?

He did not count how many syllables he could still say. He measured savings: how much less time or repetition the relearning took, as a share of the original learning. If a list took 20 minutes to learn and 12 minutes to relearn, the savings is 40%.

His savings fell quickly and then more and more slowly. After about 20 minutes, relearning took a little under 60% less work than the first learning. After an hour the saving was about 44%, after a day about a third, after six days about a quarter, and after a month still about a fifth [1][2]. Most of the drop he measured came in the first hours; after that the curve kept going down, but ever more gently.

This is where many popular versions of the curve go wrong. A saving of 44% after an hour does not mean that 44% of the list could be recited. Ebbinghaus used the saving as his measure of how much was retained, but it counts work saved, not syllables recalled. His own sentence is that after an hour, about half of the original work had to be done again before he could recite a list [1]. The popular version drops the word "work" and turns it into "you forget half of what you learn within an hour", which these experiments did not measure. Savings is also a more sensitive measure than recall: the authors of the modern replication described below note that material that seems forgotten can still be relearned faster than it was learned the first time [2].

Line chart: savings in relearning fell from 58% at 20 minutes to 21% at 31 days for Ebbinghaus, and from 47% to 4% in a 2015 replication.

Savings in relearning from 20 minutes to 31 days: Ebbinghaus (1885) and a 2015 replication

Savings is how much less time it took to learn a list of made-up syllables again than to learn it the first time, as a share of the first learning time; it is not the share of the list remembered. In Ebbinghaus's data savings were 58% after 20 minutes, 34% after a day and 21% after 31 days. In the replication they were lower at every interval, and after 31 days only 4%, a gap its authors could only speculate about. Each study had one participant. The time axis is logarithmic.

Data: Ebbinghaus (1885), ch. VII, § 28; Murre & Dros (2015), PLOS ONE 10(7): e0120644, Table 3 (CC BY 4.0). [1][2]

Data table
Savings in relearning from 20 minutes to 31 days: Ebbinghaus (1885) and a 2015 replication
Time since learningEbbinghaus (1885)Replication: Murre & Dros (2015)
20 minutes58.2%47.2%
1 hour44.2%37.3%
9 hours35.8%27.6%
1 day33.7%31.7%
2 days27.8%23.0%
6 days25.4%16.8%
31 days21.1%4.1%
  • Savings: the time relearning saved, as a share of the time the first learning took.
  • Ebbinghaus's own table gives his first three intervals as 0.33, 1 and 8.8 hours; the replication's were 20 minutes, 1 hour and 9 hours.
  • Ebbinghaus learned each list until he could recite it twice without an error; the replication, until once.
  • The replication's authors also give its 31-day value corrected for a steady rise in learning time over the experiment: 13.7%.

How much do we forget in 24 hours?

A version of the curve that circulates online says "50% in an hour, 70% in a day, 90% in a week". The experiments behind the curve did not measure that. After a day, Ebbinghaus's savings were about a third: relearning a list still took about two thirds of the work its first learning had taken [1]. That is one person learning lists of nonsense syllables, and the measure is work saved in relearning, not the share of a list he could still recall. How much a student forgets of a lecture in a day is not a number these studies give.

Has anyone repeated the experiment?

Yes, several times. The most faithful recent replication was published in 2015 by Jaap Murre and Joeri Dros [2]. As in the original there was one participant, the second author. He learned about 70 lists of 104 nonsense syllables over roughly 70 hours of sessions and relearned each one after 20 minutes, 1 hour, 9 hours, or 1, 2, 6 or 31 days.

The result closely resembled Ebbinghaus's curve: steep at first, then flattening. The savings ran somewhat lower, and much lower at 31 days. The authors could only speculate about why: this participant may simply forget more over long delays, or the 31-day lists, all learned early in the experiment, may have been learned under different conditions. Their data showed little link with how many other lists came in between [2].

They also asked whether the curve is perfectly smooth. Looking at their curve together with Ebbinghaus's and an earlier German replication with two participants, they concluded that it is probably not, but bumps up a little at about one day. In their own data the one-day value was higher than the nine-hour value, although their model fit found no clear bump for this participant alone. The authors point out that research on sleep and memory would predict a jump like this, and add that for this kind of experiment the link remains to be established; they also could not hold the time of day of their sessions fixed [2]. Ebbinghaus himself had suspected that his own one-day value was a little too high, and left the question open [1].

Is the forgetting curve real, then? Its shape is: 130 years later, the most faithful recent replication found the same fall, steep at first and then flattening [2]. Its numbers are another matter.

Both studies followed one person learning meaningless syllables. They show the shape of forgetting under those conditions very clearly, but they are not a timetable for anyone else.

Why does forgetting slow down?

Some natural processes lose the same fraction every hour, the way a radioactive substance decays. Forgetting does not look like that. When researchers fit equations to forgetting data, curves in which the rate of loss keeps shrinking fit far better [3].

John Wixted and Ebbe Ebbesen compared six candidate equations on word recall, face recognition and memory in pigeons. In every case a simple power function of time fitted best, and when they reanalysed Ebbinghaus's savings they found the same form [4]. A reanalysis of 210 published forgetting data sets reached a similar picture: four different equations, all with the same fast-then-slow shape, fitted most of the data sets about equally well and were hard to tell apart; memories of events from people's own lives were the main exception [5]. The power and the logarithmic curve are two of those four, and both fit forgetting data well, so neither can be called the formula of forgetting [5][3].

What these curves share is that the proportion lost per hour gets smaller as a memory ages [3]. Wixted links this to an old observation known as Jost's law: of two memories that are equally strong today, the older one fades more slowly. One explanation is that a fresh memory is fragile. In his account, the mental activity and new learning that follow it can interfere with it while it is still being stabilised, and it becomes harder to disrupt as it consolidates. He is careful about the logic: the shape of the curve fits this idea, but does not prove it [3]. Why forgetting happens at all is the subject of the article on why we forget.

Does reviewing change the curve?

Ebbinghaus tested this too. When he relearned the same lists on several days in a row, each day needed fewer repetitions than the day before [1]. In one comparison between two of his experiments, which he noted rested on few tests, 38 repetitions spread over three days had about the same effect on the next relearning as 68 repetitions made on the single day before [1]. That is an early glimpse of the spacing effect, which has its own article.

A well-known theory offers a reason. Robert and Elizabeth Bjork describe every memory with two strengths [6][7]. Storage strength is how well learned and connected something is. Retrieval strength is how easily you can reach it right now. In their theory, what you can recall at a given moment depends on retrieval strength, which drops with time and as the surrounding cues change, while storage strength, once gained, is assumed to stay. Higher storage strength slows the loss of retrieval strength [6].

The part of the theory that matters for reviewing is an asymmetry. A review or a successful recall adds less storage strength when the memory is still easy to reach, and more when it has become harder to reach [7]. In the theory, then, a review while everything is fresh adds little, and one after some forgetting adds more, as long as the review ends with the right answer, either recalled or seen again, and the storage strength it adds slows the next fall. The theory also assumes that recalling something strengthens it more than reading it again, the idea behind retrieval practice, which has its own article.

Illustrative model, not data: with no reviews the curve keeps falling; with reviews, the fall after each one is flatter.

Reviews and the forgetting curve: an illustrative model, not data

An illustrative model, not data: the same forgetting curve without reviews and with three reviews. After each review the material is easy to recall again, and the fall that follows is flatter, as the theory described above expects when a review ends with the right answer. The shape follows the research; the heights and the gaps are chosen for illustration. It is not measured on you, it is not a prediction, and it does not show that one review schedule beats another.

Shape: Wixted & Ebbesen (1991); Wixted (2004). Mechanism: Bjork & Bjork (1992). An illustration drawn for this guide. [4][3][6]

What the research does not say

  • It does not say how much you remember after an hour or a day. The classic numbers are savings in relearning work, from one person learning meaningless syllables. Read as "percent remembered", they say something the experiments did not measure.
  • It does not give you your own curve. How much of something stays with you depends on what it is, how well it was learned and what else you learn before and after it [5][3][6]. Learning something better raises the whole curve, but whether it also makes the forgetting slower is still debated [5]. The forgetting curve is a reason to come back to what you learn, not a set of numbers that predicts when a particular fact will slip away for you.
  • It does not settle which equation is right. Power, logarithmic and a few related curves fit forgetting data about equally well [5][3]. What is well supported is the shared shape: fast at first, then slower and slower.
  • It does not show that memories are wiped out. The replication's authors read savings as a sign that material which seems forgotten can still be relearned faster, so some of what cannot be recalled may still be there in some form [2]. That storage strength stays once gained is an assumption of the Bjorks' theory, not something measured directly [7].
  • It does not establish that sleep caused the one-day bump. The replication's authors offer sleep as a possible explanation, say that for this kind of experiment it remains to be established, and could not hold the time of day fixed [2].
  • The second figure is a model, not a measurement. Its shape follows the research, but its numbers are chosen for illustration. It does not show that one review schedule beats another: the evidence that ever-growing gaps work better than equal gaps is not there, as the article on spaced practice explains.
  • How memory works

    How a memory is encoded, stored and retrieved, how little working memory holds, and why failing to recall something does not mean it is gone.

  • Why do we forget what we study?

    Why you forget what you study: new learning can disturb fresh memories, similar ones compete, cues change. Forgotten often means out of reach, not gone.

  • Spaced repetition and spaced practice

    The spacing effect: why spreading study over days helps memory last, how far apart reviews should be, and whether ever-longer intervals beat equal gaps.

  • Retrieval practice (active recall)

    Retrieval practice, or active recall: why recalling and then checking the answer tends to beat rereading, how large the testing effect is, and its limits.

  • Why cramming feels like it works, and why it fails

    Cramming can carry you through a test that is hours away, which is why it feels efficient. What it costs weeks later, and why learners misjudge it.

  • Sleep and memory

    What studies show about sleep after learning, what an all-nighter costs your memory, and what is known about sleep on the night before an exam.

Sources

Every source the guide cites

  1. [1] Ebbinghaus, H. (1885). Über das Gedächtnis: Untersuchungen zur experimentellen Psychologie. Leipzig: Duncker & Humblot. Translated into English as: Memory: A contribution to experimental psychology (H. A. Ruger & C. E. Bussenius, Trans.). New York: Teachers College, Columbia University, 1913. book · psychclassics.yorku.ca
  2. [2] Murre, J. M. J., & Dros, J. (2015). Replication and analysis of Ebbinghaus' forgetting curve. PLOS ONE, 10(7), e0120644. replication · doi.org/10.1371/journal.pone.0120644
  3. [3] Wixted, J. T. (2004). The psychology and neuroscience of forgetting. Annual Review of Psychology, 55, 235–269. review · doi.org/10.1146/annurev.psych.55.090902.141555
  4. [4] Wixted, J. T., & Ebbesen, E. B. (1991). On the form of forgetting. Psychological Science, 2(6), 409–415. experiment · doi.org/10.1111/j.1467-9280.1991.tb00175.x
  5. [5] Rubin, D. C., & Wenzel, A. E. (1996). One hundred years of forgetting: A quantitative description of retention. Psychological Review, 103(4), 734–760. review · doi.org/10.1037/0033-295X.103.4.734
  6. [6] Bjork, R. A., & Bjork, E. L. (1992). A new theory of disuse and an old theory of stimulus fluctuation. In A. F. Healy, S. M. Kosslyn, & R. M. Shiffrin (Eds.), From learning processes to cognitive processes: Essays in honor of William K. Estes (Vol. 2, pp. 35–67). Hillsdale, NJ: Erlbaum. book chapter
  7. [7] Bjork, E. L., & Bjork, R. A. (2023). Introducing desirable difficulties into practice and instruction: Obstacles and opportunities. In C. E. Overson, C. M. Hakala, L. L. Kordonowy, & V. A. Benassi (Eds.), In their own words: What scholars and teachers want you to know about why and how to apply the science of learning in your academic setting (pp. 19–30). Society for the Teaching of Psychology. book chapter · teachpsych.org

This guide is general information about learning research for students, not medical advice.