The Ebbinghaus Forgetting Curve, Explained (and How to Beat It)

The Ebbinghaus forgetting curve is the single most replicated finding in the science of memory: without active rehearsal, we lose roughly half of newly learned information within a day, and the decay continues exponentially from there. Understanding the Ebbinghaus forgetting curve is the first step to building a study system that actually sticks.

This page walks through what Hermann Ebbinghaus discovered in 1885, the math behind memory decay, why spaced repetition flattens the curve, and how a modern app like SmartRecall turns the theory into a daily 10-minute habit.

What Hermann Ebbinghaus actually discovered (1885)

Hermann Ebbinghaus was a German psychologist who, frustrated by the lack of quantitative data on memory, decided to run a brutally controlled experiment with a single subject — himself. Between 1879 and 1885 he memorised lists of 2,300 nonsense syllables (consonant-vowel-consonant trigrams like ZOF, BIK, NUL) chosen specifically because they had no semantic meaning. By stripping away meaning, he removed the confound of prior knowledge: each syllable was a "fresh" memory trace.

He then re-tested himself at fixed intervals — 20 minutes, 1 hour, 9 hours, 1 day, 2 days, 6 days, 31 days — and measured the savings score: how much less time it took to relearn the list compared to learning it from scratch. The results, published in Über das Gedächtnis (1885), produced the now-famous downward exponential curve:

The shape is remarkable: most forgetting happens in the first hour, then the rate of forgetting itself slows. What survives the first 24 hours is comparatively durable.

Ebbinghaus' methodology has been replicated many times — most rigorously by Murre & Dros (2015), who reproduced his curve almost exactly using a modern subject and the same nonsense-syllable protocol. The curve is real.

The math: how memory decays

Ebbinghaus described the curve with a function that, in modern notation, is usually written:

R = e^(-t/S)

Where:

• R = retention (probability of recall, 0 to 1)
• t = time elapsed since learning
• S = relative strength of the memory trace
• e = Euler's number (~2.718)

A useful intuition: S is the "half-life knob". A freshly learned, never-reviewed item has a small S — maybe a few hours. After successful review, S grows, and the curve flattens. Plug in numbers and the dynamics become concrete:

Two takeaways jump out:

1. A new memory is fragile: with S = 1 day, retention crashes through 50% within 17 hours.
2. Each successful review buys exponentially more time: pushing S from 1 day to 30 days doesn't just give you 30× longer — it lets a review survive a month at the same retention level a fresh memory only reaches within hours.

This is why "cramming the night before" feels productive but produces almost nothing measurable a week later. You spent your time stretching t instead of growing S.

How spaced repetition flattens the curve

Spaced repetition exploits a second Ebbinghaus finding that gets less press: every time you successfully retrieve an item just before forgetting it, the curve resets — but at a shallower slope. Each review multiplies S. After 4–5 well-timed reviews, S is measured in months, not hours.

Visually: imagine the original curve as a steep ski slope. The first review at day 1 catches you at ~37% retention; you climb back to 100%, but now the slope is gentler. The next review at day 6 catches you at ~80% (because S is now ~7 days); you reset again and the slope flattens further. By the fourth review, the curve is nearly horizontal across a month. This is the "expanding interval" pattern.

The trick is timing. Review too early (e.g. cramming the same card 5 minutes apart) and the curve hasn't decayed enough — your brain doesn't update S. Review too late and you've forgotten too much; you're effectively re-learning, not reviewing. The sweet spot is just before forgetting — roughly when R ≈ 0.85–0.90.

Algorithms like SM-2, SM-17, FSRS and Leitner are all attempts to schedule that "just before forgetting" moment for each individual card. SM-2 (the algorithm behind Anki and SmartRecall) does it with a per-card ease factor that grows on success and shrinks on failure. See How SM-2 Works for the exact math, and SM-2 vs FSRS vs Leitner vs Anki for how the modern alternatives compare.

Building your own anti-forgetting system

You don't need an app to defeat the forgetting curve — but you do need discipline on four things. In order:

1. Atomise your knowledge. Break every concept into the smallest testable unit. A 200-word definition is not a flashcard; a single question with a single 1–10 word answer is. Atomic items are easy to review, easy to grade, and easy for your brain to consolidate.
2. Use active recall, not re-reading. The curve only resets when retrieval succeeds. Re-reading your notes feels productive but produces zero update to S. Cover the answer, force a guess, then check.
3. Honour the schedule. Review today's due items today, not tomorrow. A one-day delay on a card with S = 2 days means retention drops from 90% to 60% — you'll need a second review to recover, doubling your work.
4. Grade honestly. "I almost remembered" is not the same as "I remembered". The algorithm needs accurate signal. If you cheat upward, intervals grow too fast and the next review will catch you below 50% retention — slower than the curve itself.

Done by hand, this is the Leitner box system: physical cards in 5 boxes, promoted on success and demoted on failure. Done in software, it's Anki, SuperMemo, Quizlet's Learn mode, or SmartRecall.

How SmartRecall implements this for you

SmartRecall runs SM-2 on every card you create, automatically. You never touch the math. The app simply asks: was that recall easy, hard, or wrong? — and from that single grade it updates the card's ease factor, computes the next interval, and queues it for the day it's most likely to slip below 90% retention.

Three things SmartRecall does differently from a generic flashcard app:

• AI-generated atomic cards. Paste an article, a chapter, or a meeting note; SmartRecall splits it into atomic Q&A pairs that respect rule #1 above. No more 300-word "flashcards".
• Honest grading UI. The four review buttons (Again / Hard / Good / Easy) map directly to SM-2's q-scores. The labels are calibrated so "Good" really means you retrieved it without struggle — not you sort of recognised it.
• Daily review budget. Instead of dumping 200 due cards on you (the failure mode that kills most Anki users by month 3), SmartRecall caps the daily queue and rolls overflow forward intelligently, keeping daily review under 15 minutes for most users.

See pricing for the free and Pro tiers — the spaced-repetition engine itself is free forever.

FAQ

How fast do we forget?

According to Ebbinghaus' original 1885 data and Murre & Dros' 2015 replication, retention drops to roughly 58% after 20 minutes, 44% after 1 hour, and 33% after 24 hours for unrehearsed nonsense material. Meaningful material (definitions, vocabulary in context, concepts you can connect to prior knowledge) decays more slowly, but the shape — fast initial drop, then a long tail — is the same.

Is the forgetting curve still scientifically valid?

Yes, with caveats. The exponential shape and the order-of-magnitude numbers replicate cleanly in modern studies. What modern memory research adds is that the parameters (how fast the curve drops, how big the "savings" effect is) depend heavily on material type, prior knowledge, sleep, and emotional salience. Ebbinghaus deliberately controlled all of those out by using nonsense syllables on himself; real-world curves vary, but the principle holds.

What's the difference between the forgetting curve and spaced repetition?

The forgetting curve is a description of how memory decays without intervention. Spaced repetition is a prescription that uses the curve to schedule reviews at the optimal moment — just before each item is forgotten — so that each review extends the half-life maximally. One is the disease, the other is the treatment.

Can the curve be reset to zero?

Each successful review effectively resets R back to ~1.0 and increases S (the memory's strength). After 4–5 well-timed reviews, S is large enough that the curve is nearly flat over a month — so practically, yes, repeated successful retrieval can drive long-term retention asymptotically toward 100%. There is no single "reset to zero" event though; durability is built incrementally.

Is Anki based on the forgetting curve?

Indirectly, yes. Anki implements the SM-2 algorithm, which Piotr Wozniak designed in the 1980s explicitly to schedule reviews against an Ebbinghaus-style decay model. SmartRecall uses the same SM-2 algorithm. Newer schedulers like FSRS fit a more sophisticated three-component memory model to your actual review history, but they're all descendants of the same Ebbinghaus insight.

How long do I need to review a card before it's "permanent"?

There is no truly permanent — biology always wins eventually. But after roughly 6 successful reviews spread over 6+ months, most learners report S values of 1–2 years, meaning a single yearly review is enough to keep retention above 90%. That is what Wozniak calls the "long-term phase" of SuperMemo.

Ready to put the Ebbinghaus forgetting curve to work for you instead of against you? SmartRecall handles the scheduling math automatically — you just answer the question. Create a free account and start your first deck in under two minutes.