The Ebbinghaus Forgetting Curve: What He Measured and How to Beat It

The Ebbinghaus forgetting curve shows how quickly memory of new material fades when you do not review it: steeply in the first hours and days, then more slowly. Hermann Ebbinghaus measured it on himself in the 1880s by learning nonsense syllables. Its practical lesson is to review just before you forget, at growing intervals.

That is the short answer. The longer one is more useful, because the curve is often quoted with made-up percentages and drawn as a neat line it never was. Here is what Ebbinghaus actually did, what a modern replication found, and how to turn it into a study routine that works.

What Ebbinghaus actually did

Between 1880 and 1885, Ebbinghaus ran memory experiments with a single subject: himself. He memorised lists of nonsense syllables, consonant-vowel-consonant combinations like “WID” and “ZOF”, and tested himself after different delays. He published the results in 1885 in Über das Gedächtnis (“On Memory”). That history is summarised on Wikipedia’s forgetting curve page.

Two design choices matter:

• Nonsense syllables. He wanted material with no meaning, so prior knowledge could not help. That makes the curve a picture of memory at its weakest: no meaning, no context, no links to anything you already know.
• Savings, not recall. He did not just ask “can I still recite the list?”. He measured how much less time it took to relearn the list to perfection compared with the first time. If relearning was faster, some memory had survived, even if he could not recite a single syllable. That is a more sensitive measure than a simple test.

From those savings scores he fitted a curve: a fast drop soon after learning, then a slow tail. Wikipedia summarises the general finding as people tending to “halve their memory of newly learned knowledge in a matter of days or weeks unless they consciously review the learned material”.

Does it hold up? The 2015 replication

A single person testing himself in the 1880s is thin evidence, so it matters that the result was repeated. In 2015, Murre and Dros published a replication in PLOS ONE. One subject learned 70 lists of nonsense syllables and relearned them after the same delays Ebbinghaus used: 20 minutes, 1 hour, 9 hours, 1 day, 2 days, 6 days and 31 days.

Across most of the delays the savings scores were similar to the original. The biggest gap was at 31 days, where the new subject’s savings were much lower than Ebbinghaus’s; the authors offer possible reasons but leave it unexplained. There was one more interesting difference: the curve was not perfectly smooth. Retention at the one-day and two-day points was higher than the fitted curve predicted, a small jump upward starting at 24 hours. The authors suggest sleep may help, while noting it needs more research.

So the shape is real, and it is roughly the shape Ebbinghaus drew. Treat it as a shape, though. How fast you forget a given fact depends on how meaningful it is, how well you learned it, and how often you have met it before. Anyone quoting an exact percentage for “after one day” is quoting nonsense-syllable data, not a law about your biology notes.

What the curve tells you to do

The curve has two lessons that are easy to act on.

1. Review early, then less often

The steepest loss happens soon after learning. A first review the next day catches material while it is still mostly there, which makes the review quick and the memory stronger. Each successful review flattens the curve, so the next review can wait longer. This is spaced repetition: gaps that grow, for example one day, then three, then a week, then a few weeks.

2. Recall, do not reread

Rereading feels productive because the page looks familiar. But familiarity is not recall. Psychologists call the benefit of retrieving the answer the testing effect: Wikipedia’s summary notes that testing knowledge can produce better learning than re-reading or highlighting. Covering the answer and retrieving it from memory is harder, and that retrieval is the part that helps. Flashcards are the simplest way to force it: a question on one side, the answer on the other, and you commit to an answer before you flip.

Wikipedia’s summary points the same way: the two main ways to strengthen memory it lists are better representation, such as mnemonics, and “repetition based on active recall (especially spaced repetition)”.

A schedule you can use today

If you are reviewing on paper or in a simple app, this is a reasonable manual schedule:

Review	When	If you blank
1st	The next day	Review again tomorrow
2nd	3 days later	Back to 1 day
3rd	About a week later	Back to 3 days
4th	2 to 3 weeks later	Back to a week
5th	About 2 months later	Back to 2 to 3 weeks

The rules behind the table matter more than the exact numbers: stretch the gap after a success, shrink it after a miss, and review the item, not the whole chapter.

The problem appears at scale. With 30 items you can track this in a notebook. With 800, each on its own schedule, you cannot, and you end up rereading everything, which is the habit the curve tells you to drop.

How software schedules against the curve

This is the job spaced repetition software does. For years Anki’s scheduling was based on SM-2, an algorithm created for SuperMemo in the late 1980s. Newer schedulers estimate, for every single card, how likely you are to remember it today, and schedule the next review for when that likelihood falls to a target.

The open-source FSRS (Free Spaced Repetition Scheduler) is the best-known current approach. Its documentation centres on one setting, desired retention, defined as “the fraction of cards recalled successfully when they are due”. It calls this the most important setting and suggests 90% should work for most people. Higher retention means more reviews per day; lower means fewer reviews and more forgetting. Anki added FSRS as an option in 2023, according to Wikipedia’s Anki article.

In plain terms: instead of you guessing “a week seems right”, the scheduler predicts the moment each card is about to slip and puts it in front of you then.

Where Phosphor fits

Phosphor is a spaced repetition app built around exactly this idea. Each card is scheduled with FSRS (version 6) at a 90% retention target, and you grade every answer Again, Hard, Good or Easy so the schedule adapts to you. If you set an exam date, the target rises to 95% for the last 14 days, so the cards you would otherwise lose come back more often when it counts.

You can build cards from a lecture video or paste your own notes, and export a deck to Anki (.apkg) if you prefer to review there.

Whatever tool you use, the method is the same: questions, not rereading; short gaps first, then longer ones; and let something other than your memory keep track of the dates.

The short version

• Ebbinghaus measured forgetting on himself with nonsense syllables, published in 1885.
• A 2015 replication reproduced the curve, with a small bump at 24 hours that may relate to sleep.
• Forgetting is fastest early, so review the next day, then at growing gaps.
• Use active recall, not rereading, and let a scheduler like FSRS track the dates at scale.