The forgetting curve: what Ebbinghaus actually measured

The forgetting curve describes how retention of newly learned material falls quickly at first and then ever more slowly. Hermann Ebbinghaus measured it in 1879–80 and published it in 1885, by learning lists of nonsense syllables and timing how much faster he could relearn them later. What he actually tracked was not how much he could recall, but how much relearning time he saved: about 58% of the original effort saved after 19 minutes, about a third saved after 24 hours, and about a fifth saved after a full month.

That distinction matters, because almost every page that repeats the curve turns Ebbinghaus's savings figures into recall percentages he never reported: "you forget 70% within a day", "half is gone within the first hour". Here is what he actually did, what the numbers mean, and what has held up since.

• The forgetting curve comes from Hermann Ebbinghaus's 1885 book Memory, based on experiments he ran on himself in 1879–80.
• His measure was savings in relearning time, not how much he could recall or recognise: a subtler and more forgiving test than most summaries admit.
• The steep early drop and long, shallow tail have been replicated, most thoroughly by Murre and Dros in 2015, using the same method and the same intervals.
• Ebbinghaus himself warned that his formula was "a shorthand statement" of results "found but once", not a general law of memory.
• His own data also show the fix: spreading repetitions out over several days saved as much work as nearly twice as many repetitions crammed into one day.

What Ebbinghaus actually did

Ebbinghaus was his own only subject. Over roughly a year, he built a pool of about 2,300 nonsense syllables (meaningless three-letter combinations such as "wid" or "kaf"), mixed them by chance, and learned them in lists of thirteen. He deliberately avoided any memory tricks: "there was no attempt to connect the nonsense syllables by the invention of special associations", so that what he measured was the effect of repetition alone, not clever encoding.

He read through each list of thirteen syllables at a fixed pace until he could recite it twice in a row without error; that was his criterion for "learned". Then, after a set interval, he relearned the same list to the same criterion and timed how long it took. He ran 163 of these learn-then-relearn tests across 1879 and 1880, at intervals of about 19 minutes, 1 hour, 9 hours, 1 day, 2 days, 6 days, and 31 days.

What the curve measures: savings, not recall

The number Ebbinghaus reported at each interval was the saving: the percentage of the original learning time he no longer had to spend, because some trace of the list remained. If relearning took as long as the first learning, the saving was 0%; if it took no time at all, the saving was 100%. He put it plainly: "This saving in work is each time the measure for the amount remembered."

Notice what the table does not say. It does not say Ebbinghaus could recall only a third of the list after a day; it says relearning it took a third less time than the first time round. Savings can register even when nothing can be consciously recalled, because partial familiarity still speeds up relearning. That is a real form of retention, just not the one most retellings describe.

Why the viral percentages are wrong

The claims that circulate online (forgetting 90% of new material within hours, or 75% within a day or two) are not what Ebbinghaus reported. They read his savings numbers as if they were recall scores, then round them further for effect. His own 24-hour figure, "about one third was always remembered", already gets flattened in translation from "a third of the relearning effort was saved" to "you forget the other two thirds", which is not the same claim.

There are two further reasons to be careful with the specific percentages. First, the 1-hour and 8–9-hour values carried a time-of-day correction, because those relearning sessions fell at a different, less alert point in the day than the first learning session; Ebbinghaus applied the correction himself and called it inexact. Second, and more fundamentally, this was one person, learning meaningless syllables, in the 1880s. Ebbinghaus was explicit about the limits of what he had shown: the formula he fitted to his data has, in his words, "no other value than that of a shorthand statement of the above results which have been found but once and under the circumstances described." He was not proposing a universal law of forgetting; he was reporting one data set.

Meaningful material behaves differently, by his own account (see below), and psychology terms with context, examples, and repeated use are closer to meaningful material than to strings of nonsense syllables.

Is the forgetting curve really exponential?

Many summaries describe the curve as an exponential decay. The data itself has been re-examined more carefully than that. Wixted and Ebbesen, reanalysing forgetting data including a version of Ebbinghaus's own savings scores, found that a power function, where the rate of loss keeps slowing as time passes, fit consistently better than an exponential one. Rubin and Wenzel went further, fitting 210 published sets of retention data to 105 different two-parameter functions; a handful of forms, including power and logarithmic functions, gave the best fits, but the study could not clearly separate them from each other.

The safest description, and the one this article uses, is qualitative: retention falls fast at first and then ever more slowly, with no single named curve doing all the explanatory work.

Has the forgetting curve been replicated?

Yes, and fairly closely. Murre and Dros (2015) repeated Ebbinghaus's method on a single subject who spent about 70 hours relearning nonsense-syllable lists across 75 days, using the same intervals Ebbinghaus used: 19 minutes out to 31 days. The resulting curve matched the shape of Ebbinghaus's original closely enough that the authors called the replication successful.

One detail didn't match a smooth curve: savings ticked upward at the 24-hour mark rather than continuing to fall. The authors note that sleep-related memory research would predict something like this, but they stop short of claiming it as the explanation. In their words, "for this particular type of experiment this remains to be established." It is a genuine open question, not a settled one, so treat any confident claim that "sleep causes the 24-hour bump" as going beyond what the replication actually showed.

What slows forgetting down

Ebbinghaus's own results point to the fix, and it isn't in the popular curve at all; it's in a chapter most summaries skip. When he spread his repetitions across three days instead of cramming them into one, 38 spaced repetitions produced results as good as 68 repetitions done in a single sitting on the day before a test. Spacing out practice, in other words, did more with less effort than repeating the same amount all at once, though Ebbinghaus was careful to note this held "for only very limited conditions" in his own experiments, not as a general guarantee.

He also found that meaningful material needed far fewer repetitions than nonsense syllables in the first place: stanzas of English verse took less than half the repetitions his shortest syllable lists needed to reach the same "learned" criterion. And retention of meaningful material can hold up for a very long time once it is well learned. Bahrick (1984) followed more than 700 people who had studied Spanish in school and found that their recall of it fell for the first three to six years after they stopped studying, then stayed essentially flat for decades, with how well they'd originally learned it, not how long ago, predicting what stuck.

Between spacing repetitions out and testing yourself on them rather than just rereading, the practical response to a forgetting curve is well studied. Our guide to memorising psychology terminology covers the two study methods with the strongest evidence behind them, spacing and self-testing, in more depth than fits here.

How Psychology World helps you keep what you save

Forgetting is the default state for anything you look up once and move on from; that's true of nonsense syllables in 1880, and it's true of psychology terms today. What counts is meeting a term again before it fades. In Psychology World, you can save definitions to your own lists and drill them as flashcards, as multiple-choice questions once a list has four or more terms, or by typing the answer out yourself, and a daily word brings a saved term back into view without you having to go looking for it. The four steps of the learning loop and what each study mode does are both set out on the home page.