Learning and memory
What the forgetting curve shows, and what to do about it
Short answer
In Hermann Ebbinghaus's experiments, relearning a list 20 minutes later took about 58% less effort than learning it the first time. After a day the saving was about 34%, and after a month about 21%. Those numbers measure effort saved, not the share of the list he could recall. In later experiments, reviewing after a gap and testing yourself both left people recalling more a week or more later.
The forgetting curve comes from one man’s experiments on himself. Hermann Ebbinghaus spent 1879 and 1880 learning rows of 13 nonsense syllables, strings of letters picked so they meant nothing, usually eight rows to a session. Then he waited (20 minutes, an hour, 9 hours, or 1, 2, 6 or 31 days), learned the same rows again and measured how much less time the second round took. He ran 163 of these paired sessions and published the results in 1885. Plotted against time, they make the curve.
The curve, from his numbers
The figure uses the savings values that Murre and Dros took from Ebbinghaus’s 1880 manuscript for their 2015 paper. The second line is their replication. Dros, then 22, was its only subject, and he spent about 70 hours learning and relearning lists in sessions that ended in February 2012.
Relearning effort saved, by time since learning
- Ebbinghaus, 1879 and 1880
- Murre and Dros replication, 2015
The same numbers as a table, with one column worked out from them:
| Time since learning | Effort saved, Ebbinghaus | Relearning time as a share of the first learning, Ebbinghaus | Effort saved, 2015 replication |
|---|---|---|---|
| 20 minutes | 58.2% | 41.8% | 47.2% |
| 1 hour | 44.2% | 55.8% | 37.3% |
| 9 hours | 35.8% | 64.2% | 27.6% |
| 1 day | 33.7% | 66.3% | 31.7% |
| 2 days | 27.8% | 72.2% | 23.0% |
| 6 days | 25.4% | 74.6% | 16.8% |
| 31 days | 21.1% | 78.9% | 4.1% |
Savings from Murre and Dros (2015), Table 3. The middle column is 100% minus Ebbinghaus’s saving.
What the savings numbers measure
Savings is the share of the original learning time you no longer need the second time. Murre and Dros give a simple case: if a list took 25 rehearsals to learn and 20 to relearn, the saving is 5 out of 25, or 20%.
So Ebbinghaus’s 21% after a month means relearning took about 79% as long as the first learning. It doesn’t mean he had forgotten 79% of the list. Savings counts work, and a list you can’t recite a word of can still score above zero, because relearning it goes faster than starting cold.
Many explanations of the curve label its vertical axis “memory retained” and turn the one-day figure into “you forget two thirds of what you learn in 24 hours”. Ebbinghaus’s own summary made the slip easy: in the 1913 English translation of his book, he wrote that after 24 hours “about one third was always remembered”. He used the saving as his measure of what remained. What he measured was time.
How fast we forget, in Ebbinghaus’s data
Between 20 minutes and one day, Ebbinghaus’s saving fell from 58.2% to 33.7%, a drop of 24.5 points. Over the next 30 days it fell 12.6 points more, to 21.1%. So about two thirds of the decline he measured between 20 minutes and a month came in the first day, and 14 of those 37 points came between the 20-minute and one-hour tests.
The replication has the same shape, with lower values at most points. Its one clear bump comes between 9 hours and a day, where the saving rose from 27.6% to 31.7%. Murre and Dros note that research on sleep and memory would predict a jump like this one, but say it remains to be established for this kind of experiment. Their 31-day value, 4.1%, sits far below Ebbinghaus’s 21.1%; it’s one measurement from one person.
When to review, from the spacing studies
Ebbinghaus’s curve follows one learning session and then nothing. The spacing research asks what a second session does, and when it should come.
Cepeda and colleagues’ 2006 review pooled 839 assessments of spaced practice from 317 experiments in 184 articles. Its main pattern was that the gap between study sessions that gave the best recall grew as the time until the final test grew. Few of those experiments tested memory a day or more later.
The same five researchers’ 2008 study did. More than 1,350 people learned trivia-style facts, reviewed them once after a gap of up to 3.5 months, and took a final test up to a year later. The best gap was about 20 to 40% of the time until the test when the test was a week away, falling to about 5 to 10% when it was a year away. In days, that’s a review about 1.4 to 2.8 days after learning for a test in a week, and about 18 to 37 days after for a test in a year.
Treat those ratios as a rough guide: the study had two sessions, and the material was trivia. A smaller experiment points the same way with math. Rohrer and Taylor had 66 college students do the same practice problems either in one session or split across two sessions a week apart. On a test a week later, the split group scored 74% and the one-session group 49%.
Why testing yourself beats rereading
Spacing decides when you come back. Testing decides what you do when you get there.
In Henry Roediger and Jeffrey Karpicke’s 2006 experiments, college students read short prose passages, and then some took a recall test (write down everything you remember, with no feedback) while others read the passage again. Five minutes later the rereaders were ahead, 81% to 75%. A week later the order had flipped: the tested group recalled 56% and the rereaders 42%.
A second experiment pushed further. Students who read a passage once and took three tests recalled 61% of it a week later. Students who read it four times recalled 40%, and they were the group most confident they’d remember it.
Rowland’s 2014 meta-analysis pooled 159 effect sizes comparing testing with restudying and found an average advantage for testing of g = 0.50. That’s a medium effect: the average person who was tested scored higher than about 69% of the people who restudied. In studies that gave feedback on answers, the average was g = 0.73 (about 77%). Most were lab studies, and results varied widely.
On the curve above, that payoff sits in the flat part, a week after learning. Each Scroll lesson ends with one quiz question with four options, which puts the first test seconds after reading; the spacing studies are about the next look, days or weeks on.
Scroll lessons are short introductions for general interest. They aren’t medical, psychological or financial advice, and we haven’t tested whether reading them changes anyone’s memory or focus. For a decision about your health or money, talk to a qualified professional.
The Feynman Technique
Nobel Prize-winning physicist Richard Feynman's learning method: pick a concept, explain it as if teaching a 12-year-old, identify gaps where your explanation breaks down, return to the source and fill the gaps, then simplify further. The technique works because it forces active retrieval and exposes the difference between recognising something and actually understanding it. If you cannot explain it simply, you do not understand it.
Source Gleick, J. (1992). Genius: The Life and Science of Richard Feynman.
Quick quiz
You are using the Feynman Technique to learn a concept. During your explanation to an imaginary 12-year-old, you become confused at a key step. What should you do next?
Where spaced repetition comes in
Spaced repetition joins the two findings. You test yourself on an item, usually with a flashcard, and the gap before it comes back grows each time you answer correctly and shrinks when you miss. Dunlosky and colleagues’ 2013 review of 10 study techniques rated only two as high utility, practice testing and spacing out study, and this method uses both.
Growing gaps fit Cepeda’s pattern. The exact schedules in flashcard apps are design choices, though, because the 2008 study tested a single review. The spaced repetition schedule uses the 2008 ratios as a starting point for working out review dates from the date you need something by.
A routine for one thing you want to keep
- Right after you read or hear it, close the source and write down what you remember.
- Check what you wrote against the source and fix the gaps. In Rowland’s analysis, studies with feedback showed the larger effect.
- Decide when you’ll need it. If that’s a week away, look again after about 1.5 to 3 days; if it’s a year away, after about 2.5 to 5 weeks.
- At each review, test yourself first and reread only what you missed.
For books and articles, how to remember what you read applies the same steps chapter by chapter. To try it here, close this page and write down the savings Ebbinghaus measured after a day and after a month. Then come back and check.
Drafted with AI assistance and checked against the sources listed on October 17, 2026.