In 1885, Hermann Ebbinghaus published a memory study unlike anything before it - not because the question was new, but because he insisted on measuring it precisely, on himself, with material stripped of any prior association. The result is one of the most replicated findings in psychology, and it's still the clearest single explanation for why "I read it, I should remember it" so reliably turns out to be false.
In short
- Ebbinghaus memorized nonsense syllables and measured how much effort it took to relearn them after different gaps. Loss was fast at first, then slower.
- The shape has been replicated, including in a careful 2015 repeat of his procedure.
- The curve is not a flaw in you. It is the default behaviour of memory.
- Rereading barely changes it. Successfully retrieving the information, at a well-chosen moment, does.
- The practical problem is timing: estimating, per item, when you are about to forget.
What Ebbinghaus actually measured
Ebbinghaus memorized lists of nonsense syllables - meaningless three-letter combinations such as "WID" or "ZOF", chosen specifically so he couldn't lean on existing knowledge to remember them - then tested himself after gaps ranging from twenty minutes to thirty-one days. His self-study ran from 1880 to 1885, and the results were published as Über das Gedächtnis.
His measure is worth understanding, because it isn't what most people assume. He didn't simply ask "can I recite the list?" He used what's called the savings method: how much less time it took to relearn a list the second time compared with the first. If relearning took half the original effort, that was 50% savings. It's an indirect but sensitive measure, because it can detect memory traces too faint to produce a successful recall (overview of the forgetting curve).
Plotted as savings against time elapsed, the result is the curve that bears his name. It isn't gentle. In his data, a large part of the benefit was gone within the first hour, and more of what remained disappeared over the following day. After that first steep drop, the rate of loss slows down substantially - what survives the first day tends to stick around longer - but it never flattens to zero. Left alone, memory keeps leaking, just more slowly over time.
A picture you can keep in your head
Imagine learning ten new words on a Monday morning. By Monday evening you can still produce most of them, though a few feel slippery. By Wednesday you recognise most when you see them but can produce far fewer. By the next Monday, several are fully gone and a few are surprisingly solid. The ones that survived are usually those you used, were curious about, or happened to meet twice.
None of that is mysterious, and it isn't a character flaw. The steepness varies by material and by person, but the pattern - quick early loss, long slow tail - is the same.
Why it still holds up
You'd expect a 140-year-old self-experiment on one subject to have been superseded by now, and in the narrow sense that better-controlled, larger studies followed, it has. But the core shape - fast initial loss, slowing decay, no natural floor - has been replicated across a wide range of material and populations since: word lists, faces, foreign vocabulary, textbook facts. The exact steepness varies by how meaningful the material is to you and how it was learned, but the shape of the curve itself is remarkably durable.
There is even a direct repeat of his procedure. In 2015, Murre and Dros ran the savings method with nonsense syllables again, with intervals from twenty minutes to thirty-one days, and concluded that the Ebbinghaus forgetting curve had indeed been replicated. They also saw something he hadn't highlighted: the curve was probably not perfectly smooth, with a small upward jump from the 24-hour mark (the paper). It's a reminder that real data is bumpier than a textbook diagram, while the broad shape survives.
That durability is what makes it useful. It's not a description of forgetful people. It's a description of how memory consolidates by default, for everyone, absent any intervention.
The forgetting curve isn't a flaw in your memory. It's the baseline your memory runs on unless something interrupts it.
What the curve does not say
A few caveats keep the idea honest.
- It's an average shape, not a prediction for a specific fact. Meaningful, connected material is retained better than arbitrary strings. A sentence that matters to you decays differently from a random code.
- Nonsense syllables are the harshest case. Ebbinghaus used them deliberately because they strip out meaning. Real-world learning usually has more hooks, so real forgetting is often slower, though still real.
- It doesn't mean you must memorize everything. The point isn't to turn your whole life into flashcards. It's to be aware that anything you want to keep needs a deliberate return trip.
Three common misreadings
"I forget half of everything in an hour." That figure comes from nonsense syllables and the savings measure. It is a statement about the harshest kind of material, not a law that applies to a conversation you cared about.
"The curve means my memory is bad." The curve describes everyone, including people with excellent memories. What differs between people is the steepness and how much prior knowledge they can hang new material on.
"If I just review more, the curve goes away." More exposure in one sitting changes little. The gaps matter: the same ten minutes spread across several days does much more than ten minutes in one block. That is the spacing effect (overview), and it is why timing is the whole game.
The one thing that reliably bends it
Rereading something doesn't bend the curve much - it mostly just delays the next drop by a small amount, and it creates a sense of familiarity that is easy to mistake for knowing. What actually changes the shape of the curve is retrieval: successfully pulling the information back out of memory yourself, under some effort, rather than having it placed back in front of you. Research on this is usually filed under the testing effect (overview), and it's the reason flashcards ask a question before showing the answer. The difference between the two ideas - how you review versus when you review - is the subject of active recall vs spaced repetition.
Do that at the right moment - close to when you'd otherwise have forgotten, not right after you learned it - and the next curve for that item starts from a flatter, slower-decaying point than the first one did. Repeat it enough times at the right intervals, and an item that needed daily reinforcement early on can go weeks or months between reviews and still hold. That's the whole mechanism behind spaced repetition.
What "well-timed" actually means
The hard part was never knowing that spaced review works - Ebbinghaus effectively showed that in the 1880s. The hard part is knowing exactly when "about to forget" is, for this specific item, for you, today. Too early and you're reviewing something you already know solidly, wasting time. Too late and you've already lost it, which means relearning instead of reinforcing - a meaningfully different (and more expensive) task.
Early systems answered with a schedule you could run by hand. The box-and-card approach in the Leitner system is a good example: right answers move a card to a box reviewed less often, wrong answers send it back to the start. Later systems tried to estimate your memory more directly. In the model behind FSRS, the forgetting curve is not a diagram but an input: each card has a stability value and a predicted chance of recall at any moment, and the scheduler picks the review time that matches a target.
What to do with this
You don't need the equations to use the lesson.
- Treat the first day as the dangerous one. If something matters, plan a return trip soon after you learn it, then a longer gap, and so on.
- Retrieve instead of reread. Close the page and try to say it before looking.
- Pick fewer, better things. Not everything deserves a place in rotation. How to write good flashcards is about making what you do keep as small and clear as possible.
- Use a tool that handles the timing. Cirello schedules with FSRS-6 so you aren't counting days yourself; see the FSRS app page for how.
That's the actual engineering problem behind any serious spaced-repetition system: not proving the curve exists, but estimating, continuously and per item, where you currently sit on it. Get that timing close to right and review stops feeling like a chore you're behind on - it starts landing exactly when the item was about to need it.