Forgetting Curve: What Ebbinghaus Actually Found

What Is the Forgetting Curve?

The forgetting curve is a graph of how much of something you've learned stays in memory over time. It falls steeply in the first minutes and hours after learning, then flattens out, so whatever survives the first few days fades much more slowly. The German psychologist Hermann Ebbinghaus described it first, in his 1885 book Über das Gedächtnis (published in English in 1913 as Memory: A Contribution to Experimental Psychology), based on experiments in which he was the only subject.

Three ideas sit inside that shape:

• Most forgetting happens early. The biggest losses come in the first hour and the first day.
• The rate of forgetting slows down. A memory that has lasted a week is far more stable than one that's an hour old.
• Reviewing changes the curve. A well-timed review, especially one where you pull the answer out of memory, slows the next round of forgetting.

What the curve doesn't give you is a universal percentage. Ebbinghaus's curve describes one man memorizing meaningless syllables, scored with a measure most summaries never explain. The numbers that circulate online ("half of it gone in an hour," "70% in a day") come from that narrow setup, and they don't transfer cleanly to a textbook chapter, a lecture or an article you cared about.

Ebbinghaus fit his data with a logarithmic formula. Later researchers found that a power function describes most forgetting data at least as well. In practice they say the same thing: steep, then flat, with a long tail.

---

The Ebbinghaus Forgetting Curve Experiment

Ebbinghaus wanted to study memory without the mess of meaning. Poetry and prose, he wrote, bring in "a multiplicity of influences which change without regularity and are therefore disturbing": associations, shifting interest, lines that stick because they're striking or beautiful. His fix was the nonsense syllable: a consonant, a vowel and a consonant, strung into something pronounceable but meaningless (think "fep" or "gim"). He built about 2,300 of them, mixed them up and drew lists at random.

The experiment behind the famous curve ran in 1879 and 1880:

1. Learn. He studied lists of 13 syllables until he could recite them twice without an error, timing the whole process.
2. Wait. A fixed interval passed: 20 minutes, 1 hour, about 9 hours, 1 day, 2 days, 6 days or 31 days.
3. Relearn. He learned the same lists again to the same standard, and timed that too.
4. Compare. The time saved on relearning, as a percentage of the original learning time, was his measure of what memory remained.

He logged 163 of these learn-and-relearn tests, each usually covering eight lists. He defined his measure, savings, as "the saving of work evident in relearning, the equivalent of the amount remembered from the first learning expressed in percentage of the time necessary for this first learning."

Here's what he found:

In his own words: "One hour after the end of the learning, the forgetting had already progressed so far that one half the amount of the original work had to be expended before the series could be reproduced again; after 8 hours the work to be made up amounted to two thirds of the first effort. Gradually, however, the process became slower so that even for rather long periods the additional loss could be ascertained only with difficulty."

That last sentence sums up the curve. Between 20 minutes and one day, savings fell by about 25 points. Between day 1 and day 31, they fell by about 13.

The savings method has a hidden strength: it can detect memories too faint to recall. A list you can't recite a single syllable of may still be quicker to relearn, which shows something survived. It's also why savings percentages aren't the same thing as "how much you remember," the confusion behind most of the myths below.

---

Five Myths About the Forgetting Curve

The forgetting curve is one of the most-cited findings in psychology and one of the most mangled. Training decks, study apps and productivity blogs repeat versions of it that Ebbinghaus wouldn't recognize.

Myth 1: The curve tells you what percentage you remember. Ebbinghaus never counted how many syllables he could recite after a delay. He measured how long relearning took. A savings score of 33.7% after one day means relearning took about two-thirds as long as the first time. It doesn't mean he could recite a third of the list. Recall can sit near zero while savings stay well above it.

Myth 2: You forget 70% of everything within a day. This one is everywhere in workplace training material. It's roughly 100% minus Ebbinghaus's one-day savings score, which makes it a statement about relearning time rather than recall, and it rests on one person learning syllables built to be meaningless. His own tests with Byron's poetry, covered below, found stanzas far quicker to learn and easier to relearn a day later than his short syllable lists.

Myth 3: Forgetting is exponential. Many explainers present the curve as R = e^(-t/S), a tidy exponential decay. It isn't Ebbinghaus's formula, and it isn't the best fit.

In 1991, John Wixted and Ebbe Ebbesen ran three experiments (word recall, face recognition and a memory task for pigeons) and compared six candidate functions. A simple power function won, and it also fit a reanalysis of Ebbinghaus's own savings data.

Five years later, David Rubin and Amy Wenzel fit 105 different two-parameter functions to 210 published data sets. The best fits were the logarithmic function, the power function, the exponential in the square root of time and the hyperbola in the square root of time. Plain exponential decay wasn't on the list.

A power curve has a long, fat tail: once a memory has lasted a while, it decays far more slowly than an exponential would predict. Old knowledge is sturdier than the scary charts suggest.

The last two myths (one curve for everyone, one schedule for everything) get their own sections below.

---

Is the Forgetting Curve Real? What the Replications Found

Yes. The basic shape has replicated, which is rarer in psychology than it should be. In 2015, Jaap Murre and Joeri Dros of the University of Amsterdam published a careful replication in PLOS ONE.

Dros, the second author, was the only subject, just as Ebbinghaus had been. Because he was a native Dutch speaker, the authors called it the first non-German replication. Between December 2011 and February 2012 he spent about 70 hours learning lists of 104 nonsense syllables (eight rows of 13, as many as Ebbinghaus learned in one eight-list session) and relearning them after the same seven intervals.

The authors concluded "that the Ebbinghaus forgetting curve has indeed been replicated and that it is not completely smooth but most probably shows a jump upwards starting at the 24 hour data point." Their comparison also included two subjects, Mack and Seitz, from an earlier German replication by Heller, Mack and Seitz (1991):

Four people, more than a century apart, produced the same basic shape. Two details stand out.

The one-day bump. In Dros's data, savings were higher after one day (31.7%) than after nine hours (27.6%), and Mack's data show the same rise. Ebbinghaus had seen a milder version: his savings barely moved between 9 and 24 hours (35.8% to 33.7%), then dropped faster over the next day.

Ebbinghaus considered whether "night and sleep" could account for the pattern, decided they couldn't, and concluded his 24-hour figure was probably "somewhat too large." He still added, "I am in doubt about it," because tests from a separate study in 1883 and 1884 gave almost the same number (33.4%).

Murre and Dros saw the bump differently. "Current research on the effects of sleep on memory would predict such a jump," they wrote, while noting that for this kind of experiment it "remains to be established." A nine-hour gap can pass within a single waking day; a one-day gap always includes a night's sleep. In their curve fits, the jump was largest for Mack, small for Ebbinghaus and absent for Dros, whose fit they thought was probably skewed by his very low one-month score.

The 31-day outlier. Dros's savings after a month were just 4.1%, far below everyone else's. The authors traced part of that to a steady creep in his learning times over the course of the experiment, which they said may be due to proactive interference (earlier lists getting in the way of new ones) or fatigue. Correcting for the creep raised the figure to 13.7%, still well below the other three, whose month-later savings ran from 20.1% to 25.8%.

The shape holds well beyond nonsense syllables, too. In Rubin and Wenzel's survey of a century of retention studies, the same small family of curves fit most of the 210 data sets, with autobiographical memory as the main exception.

---

What Changes the Shape of Your Forgetting Curve

Ebbinghaus picked nonsense syllables to keep meaning out of his data, which also made them close to a worst case: nothing to hook into. Real learning rarely looks like that, and several factors bend the curve.

Meaning. Ebbinghaus ran a side experiment in 1884 with stanzas from Byron's Don Juan. Each 80-syllable stanza took him "hardly nine repetitions" to learn; projecting from his syllable data, he estimated that 80 to 90 nonsense syllables would have needed 70 to 80. He called it "the extraordinary advantage which the combined ties of meaning, rhythm, rhyme, and a common language give to material to be memorised." The stanzas also held up well. A day later, relearning them took about half the original work, while his 12-syllable lists took about two-thirds, and the stanzas had needed fewer than half as many repetitions as those lists to learn in the first place. (His much longer 24- and 36-syllable lists, which took far more repetitions to learn, did as well or better.)

How well you learned it in the first place. Harry Bahrick's 1984 study tested 733 people on the Spanish they had studied in school, up to 50 years earlier. Their knowledge declined for the first three to six years, then stayed essentially unchanged for up to 30 years before a final decline.

"Large portions of the originally acquired information remain accessible for over 50 years in spite of the fact the information is not used or rehearsed," Bahrick wrote. He called that durable layer the "permastore." Its size tracked the level of original training (how much Spanish people had taken and the grades they'd earned), and most of them had barely practiced since.

Sleep. Two students took part in John Jenkins and Karl Dallenbach's 1924 study: they learned nonsense syllables and were tested after 1, 2, 4 or 8 hours spent either asleep or awake. They remembered more after sleeping. Jenkins and Dallenbach concluded that forgetting "is not so much a matter of the decay of old impressions and associations as it is a matter of the interference, inhibition, or obliteration of the old by the new." Modern research treats sleep as an active period for consolidating memories, which fits the one-day bump in the replications.

Interference. New learning competes with old learning, and old learning gets in the way of new. It's one possible reason Dros's learning times crept up over his experiment, though Murre and Dros couldn't pin it down.

How you review. Rereading and recalling feel similar but produce very different curves. That's the subject of the next section.

---

How to Beat the Forgetting Curve

You can't stop forgetting, and you wouldn't want to. But two techniques reliably make the curve flatter after each review: spacing your reviews out (spaced repetition) and making each review a test (retrieval practice).

Spacing. Ebbinghaus found the spacing effect too. Comparing two of his experiments, he found that 38 repetitions spread over three days left a 12-syllable list as well learned as 68 repetitions crammed into a single day. That, he wrote, "makes the assumption probable that with any considerable number of repetitions a suitable distribution of them over a space of time is decidedly more advantageous than the massing of them at a single time."

How far apart should reviews be? The most useful answer comes from Nicholas Cepeda, Edward Vul, Doug Rohrer, John Wixted and Hal Pashler (2008). They taught 1,354 people a set of obscure facts, gave them one review after a gap of up to three and a half months, and tested them up to 350 days after that review. The best gap depended on how far away the test was:

Getting the gap right mattered. Compared with reviewing straight away, the best gap they tested improved final recall by 10% for the one-week test and by 59%, 111% and 77% for the longer ones.

Two practical rules fall out of the study. The further away you need the material, the longer the gap, though it shrinks as a share of the total. And gaps that are a bit too long cost less than gaps that are too short: performance rose quickly as the gap grew toward the best value, then fell off only gradually beyond it.

Retrieval. Henry Roediger and Jeffrey Karpicke (2006) had college students read two short prose passages, one about the sun and one about sea otters. Each student reread one passage and took a recall test, without feedback, on the other.

Five minutes later, rereading looked better: 81% recalled versus 75%. Two days later it had flipped (68% for the tested passage versus 54%), and after a week the tested passage stood at 56% against 42%. So the tested passage was remembered as well after a week as the reread one had been after two days.

Their second experiment was starker. Students who studied a passage for a single five-minute period and then took three recall tests in a row remembered 61% of it a week later. Students who spent all four periods, back to back, rereading remembered 40%, even though they had read it 14.2 times on average against 3.4.

Rereaders tested after a week recalled 52% less than rereaders tested after five minutes; for the repeat testers, the drop was 14%. And the rereaders were the ones who predicted they'd remember it best.

Those are the two techniques Dunlosky and colleagues rated "high utility" in their 2013 review of ten study techniques: practice testing and distributed practice. Highlighting and rereading were rated low. Our guide to active recall covers the testing side in depth.

---

Spaced Repetition Schedules: Fixed, Expanding or Adaptive?

If spacing works, which pattern of spacing works best? Schedules come in three broad families.

Fixed schedules put every item on the same dates. The 2357 method, for example, has you learn a topic on day 1 and review it on days 2, 3, 5 and 7. They're easy to plan, which is why they're popular before exams.

Expanding schedules stretch the gap each time you get something right. The Leitner system is the classic paper version: cards climb into boxes you review less and less often, and a miss sends a card back to the start.

Does expanding beat equal spacing? Not reliably. A 2021 meta-analysis of 29 studies by Alice Latimier, Hugo Peyre and Franck Ramus found a large benefit for spacing retrieval practice over massing it (g = 0.74), but essentially no difference between expanding and uniform gaps (g = 0.034).

Adaptive schedulers model your own forgetting curve and bring each item back just before it's likely to slip. Anki has supported the FSRS algorithm since version 23.10. It fits a power-law forgetting curve to your review history and gives each card a "stability," the number of days until your predicted chance of recalling it falls to 90%. Reviews land when that chance hits your "desired retention," which is 90% by default.

Duolingo built its own model. Burr Settles and Brendan Meeder's 2016 half-life regression estimates the "half-life" of each word in a learner's memory, and it predicted recall with at least 45% less error than the Leitner-style algorithm Duolingo had been using.

So the schedule matters less than doing the reviews at all. Any plan that spaces them out and makes you retrieve the answer should beat rereading whenever you happen to remember.

---

Using the Forgetting Curve on What You Read and Watch

Most of what we read rides the steep part of the curve, because we never come back to it. A highlight feels like progress, but it's a pointer to a future review, not the review itself. Dunlosky's review rated highlighting as low utility. A highlight earns its keep when it feeds the two techniques that do work.

Here's a routine built on the research above:

1. Highlight less, and add a line in your own words. Use Glasp's web highlighter to mark the few sentences you'd want to be tested on, and note why each one matters. Putting it in your own words forces you to process the meaning, the opposite of a nonsense syllable.
2. Test yourself before you close the tab. In Roediger and Karpicke's studies, the first recall test came within minutes of reading. Close the page and write down what you remember; the blurting method is a structured way to do it. Then check what you missed against your highlights.
3. Let your highlights come back to you. Glasp's Daily Highlight Review email sends you a random handful of your past web highlights every day, twice a week or once a week; you pick the frequency in your email settings. Treat each one as a prompt: before you reread it, try to recall the context and why you saved it.
4. Do the same for video. Lectures fade too. YouTube Summary gives you a timestamped transcript and summary, so you can highlight the lines worth keeping and review them alongside your reading.
5. Bring your books into the loop. Import your Kindle highlights and set a frequency for the Daily Kindle Review email, so book passages resurface too.
6. Quiz yourself across everything. Write down your answer to a question first, then ask Glasp's Chat with Highlights what your highlights say about it. Its replies point back to the passages they came from, so you can check your answer against what you saved.
7. Hand the must-keep facts to a scheduler. For vocabulary, formulas and definitions you'll need for months, turn the highlight into a flashcard and let a Leitner box or Anki handle the timing.

---

Frequently Asked Questions

What is the forgetting curve in simple terms?

The forgetting curve shows how quickly we lose information after learning it. Memory drops fast in the first hours and days, then the decline slows down. Hermann Ebbinghaus first described it in 1885, and reviewing material, especially by recalling it from memory, makes the curve flatter.

How fast do we forget what we learn?

There's no single number. In Ebbinghaus's experiments with nonsense syllables, relearning a list a day later took about two-thirds of the original time, and a month later nearly four-fifths. Meaningful material is usually measured by recall instead: in Roediger and Karpicke's 2006 study, students who reread a prose passage recalled 40% of its ideas a week later, while students who had practiced recall remembered 61%.

Is the forgetting curve real?

Yes. The basic shape, steep at first and then flatter, has held up for well over a century, most directly in Murre and Dros's 2015 repeat of Ebbinghaus's experiment. What doesn't hold up are the precise percentages people attach to it, which come from one person learning meaningless syllables.

Is it true that we forget 70% of what we learn in 24 hours?

No. That figure is roughly 100% minus Ebbinghaus's one-day savings score of 33.7%, and savings measured how much faster he could relearn a list of nonsense syllables, not how much he could recall. It doesn't apply to everything you learn, and how much you keep depends heavily on meaning, sleep and how you review.

How do you beat the forgetting curve?

Combine retrieval and spacing. Test yourself instead of rereading (close the book and recall, answer flashcards, explain it out loud), and spread those tests out, with longer gaps for material you need to keep longer. Sleep helps too. A quick self-test right after learning, then a few spaced reviews before you need the material, is a sensible default.

What is the best spaced repetition schedule?

It depends on when you need the material. Cepeda and colleagues (2008) found the best gap before a review was about 3 days for a test a week later and about 27 days for a test nearly a year later. Expanding gaps haven't reliably beaten equal ones, so pick a schedule you'll actually keep: a fixed plan before an exam, a Leitner box for flashcards, or an adaptive app like Anki for large decks.

Is there a formula for the forgetting curve?

Ebbinghaus fit his data with b = 100k / ((log t)^c + k), where b is the percentage saved on relearning, t is the time in minutes (counted from one minute before learning ended), log is the base-10 logarithm, k = 1.84 and c = 1.25. The exponential R = e^(-t/S) that many websites show isn't his equation. Wixted and Ebbesen (1991) found that a simple power function beat five alternatives, including the exponential and logarithmic ones, and Rubin and Wenzel (1996) found the logarithmic and power functions among the four best fits. Anki's FSRS scheduler uses a power curve.

Who discovered the forgetting curve?

Hermann Ebbinghaus (1850-1909), a German psychologist, described it in Über das Gedächtnis (1885), translated into English in 1913 as Memory: A Contribution to Experimental Psychology. He was his own and only subject, memorizing and relearning lists of nonsense syllables over several years.

---

Schedule Your First Review Before You Forget

The forgetting curve describes what happens to anything you learn once and never revisit. If yours is steep, look at how you learned and reviewed the material before you blame your memory. Ebbinghaus showed how steep the early drop is, and the research since has shown how much of it you can avoid: make each review a test, and space your reviews to match how long you need to remember.

Start with whatever you read today. Highlight the few ideas worth keeping with Glasp, test yourself before you close the tab, and let your highlights come back to you in a review email. The steepest drop comes early, so put your first review on the calendar now. Glasp's free spaced repetition calculator turns a start date, and an exam date if you have one, into review dates you can download as a calendar file.