The Forgetting Curve: Ebbinghaus's Real Numbers and How to Flatten It

The forgetting curve at a glance

• Who: Hermann Ebbinghaus, German psychologist, published in 1885 as Über das Gedächtnis (“On Memory”).
• Shape: a fast drop at first, then a long, slow tail. Most of the loss happens in the first day.
• What he measured: “savings”, the share of learning time saved when relearning a list, not the share he could still recite.
• His numbers: 58% saved after 20 minutes, 34% after a day, 21% after a month.
• How to flatten it: review before most is lost, test yourself instead of rereading, and space the reviews out.

How Ebbinghaus measured forgetting

Ebbinghaus wanted to study memory without the help of meaning, so he built lists of nonsense syllables (consonant–vowel–consonant strings such as “DAX” or “BOK”) and learned them himself, reading each series aloud at a fixed pace until he could recite it without error. Each test session used eight series of 13 syllables. After a delay he relearned the same series and timed how long it took.

The time saved on relearning, as a percentage of the original learning time, was his measure of what remained. He called it savings. Between 1879 and 1880 he ran 163 of these paired sessions at seven delays, from about 20 minutes to 31 days.

The original numbers

Figures from Table 3 of Murre & Dros (2015), which reprints Ebbinghaus’s values. As Ebbinghaus summarised it, forgetting was very rapid at the beginning and very slow at the end: after an hour more than half of the original work had to be done again, after a day about two-thirds, after a month about four-fifths. He fitted a formula to the data (savings = 100k / ((log t)^c + k), with k = 1.84 and c = 1.25, t in minutes) and insisted it was only a shorthand for one person’s results under his conditions.

The 70% myth: what savings does and does not mean

Most popular versions of the curve say we forget about 50% within an hour, 70% within a day and 90% within a week. Those figures do not come from Ebbinghaus. They come from reading his savings scores as if they were the share of the material still remembered, and then rounding down further for the week.

Three things are wrong with that reading:

1. Savings is not recall. A list you can no longer recite still leaves a trace that makes relearning faster. Savings of 34% means relearning took a third less time, not that a third of the syllables came back.
2. The material was meaningless on purpose. Nonsense syllables have nothing to hang on to. Material you understand, that links to what you already know, is forgotten more slowly.
3. He learned to the bare minimum. Each list was learned only until it could be recited without error. Learning beyond that point, or learning it on more than one occasion, slows forgetting.

The shape of the curve is real and robust. The headline percentages are not a law of memory, and nobody should plan their study around them.

The 2015 replication

Jaap Murre and Joeri Dros at the University of Amsterdam repeated the experiment as closely as they could. One subject, aged 22, spent about 70 hours over 75 days learning and relearning 70 lists of 104 nonsense syllables (eight rows of 13), at the same seven delays. The resulting curve closely resembled Ebbinghaus’s and two earlier replications. The main exception was the 31-day point, much lower for Dros; the authors note that his 31-day lists were all learned in a compressed early block of the study.

One detail survived the replication. In Ebbinghaus’s data and in two of the replications, savings at one day is a little higher than a smooth curve predicts, a small “jump”. Ebbinghaus noticed it and repeated the one-day measurement to be sure. Murre and Dros conclude that the curve “is not completely smooth” and point to sleep as a likely reason, while noting that this has not been proven for this kind of experiment.

Is the forgetting curve exponential?

Many articles describe the curve as exponential decay. That is not quite what the data show. John Wixted and Ebbe Ebbesen (1991) compared six mathematical forms on word recall, face recognition and a pigeon memory task, and in all three a simple power function described forgetting best; their reanalysis of Ebbinghaus’s savings data found the same. A power curve falls faster than an exponential at first and much more slowly later on, which matches the long tail in the table above: what survives the first few days tends to last.

Murre and Dros fitted several functions to all four data sets. Ebbinghaus’s own formulas explained 99.8% of the variance in his data, and a power function fitted all four sets well. A sum of two exponentials, which the authors read as memory moving from a short-lived store into a more durable one, fitted about as well; the differences between the leading candidates were small.

Why the curve drops at the start

Right after learning, a memory has not yet been consolidated, the process by which the brain stabilises new memories into long-term memory. Several processes act on it in that window: new learning interferes with the old, cues present at learning are no longer around, and weak traces simply fade. Memories that make it through consolidation are much harder to lose, which is why the curve flattens.

Sleep is part of that story. In 1924 John Jenkins and Karl Dallenbach found that two students recalled more nonsense syllables after hours of sleep than after the same hours awake. A 2010 review in Nature Reviews Neuroscience by Susanne Diekelmann and Jan Born describes how sleep actively consolidates new memories, which fits the bump at one day in the forgetting-curve data.

How to flatten the forgetting curve

Every well-timed review resets the curve, and each new curve is shallower than the last. That is the idea behind spaced repetition, and the research supports four practical rules:

1. Review early. The steepest loss is in the first day, so the first review belongs within a day or two of learning.
2. Test, do not reread. Henry Roediger and Jeffrey Karpicke (2006) had students either restudy prose passages or take recall tests on them. Restudying won on a test five minutes later; testing won clearly on tests two days and a week later, even though the restudy group felt more confident.
3. Space the reviews out. A 2006 meta-analysis by Nicholas Cepeda and colleagues pooled 839 comparisons from 317 experiments: spaced study beat massed study, and the best gap grew with how long the material had to last.
4. Make it meaningful. Link new material to what you know, and for hard lists use mnemonic devices; the meaningless syllables Ebbinghaus chose are the worst case, not the normal one.

Timetables such as the 2-7-30 rule (review after 2, 7 and 30 days) or 24 hours, 7 days and 30 days are reasonable ways to put these rules into practice. Their particular numbers have not been tested against alternatives; the evidence is for spacing and self-testing, not for one fixed schedule.

The forgetting curve and IQ

Forgetting fast is normal. Everyone’s curve drops steeply at the start, and how fast a particular list fades depends mostly on how it was studied and reviewed, not on how intelligent the learner is. The memory ability most closely tied to reasoning is working memory, the capacity to hold and manipulate a few items for seconds, which is part of most IQ tests. Reducing forgetting with better study habits keeps more of what you learn, which builds the knowledge side of intelligence described in our article on fluid vs crystallized intelligence, but it does not raise general reasoning ability.

Trust and scope notes

This page summarises published research for general information. The savings figures are taken from Table 3 of Murre & Dros (2015), which reproduces Ebbinghaus’s values. It is not medical advice: forgetting that is new, worsening or affecting daily life deserves a conversation with a doctor, and no online test can assess it. The IQ Revealed test is an online reasoning test of 40 original items for adults aged 18 and over; it is not a clinical assessment and it does not measure memory. Our guide to how IQ tests work explains what an online test can and cannot tell you.

IQ Revealed is not affiliated with, endorsed by or connected to the researchers, universities, journals or publishers named on this page. Names are used only to identify the research being described.