What Is the Forgetting Curve? How Much Vocabulary You Keep

How much of what you learn today will you still know a month from now?

We ran the numbers on anonymous Tomaru learning data. The result: words reviewed on schedule are about 80% remembered when tested the next day, and 80–90% remembered after a week to a month. But a new word you didn't know the first time, left for a week before its first review, comes back less than half the time.

This article explains what the forgetting curve is, and what these numbers actually mean for learning vocabulary.

What the forgetting curve is

The forgetting curve describes how memory decays over time. It was proposed by the German psychologist Hermann Ebbinghaus in 1885. Its shape has two features: the loss is steepest in the first few days, then it flattens out.

This curve is the theoretical basis for spaced repetition. If memory decays at predictable points in time, review it once before it falls past the point of retrieval, and the curve gets pushed back up.

Ebbinghaus measured how much you forget when you never review. We are filling in the other half: how much memory actually survives when you do.

Finding 1: on schedule, 80–90% are still there after a week to a month

In Tomaru, when a word comes back next depends on how well you know it right now: words you are still shaky on come back the next day, words you know well come back after a week or a month.

The gap gets longer, yet the share remembered does not fall, because only words you already know get pushed out to long gaps, and each review is timed just before you would forget. That is what spaced repetition does: it pulls the curve back up before it drops.

In practical terms: review on schedule and only one or two words in ten slip away.

Finding 2: leave a new word a week before reviewing it, and you keep less than half

Now look at words you have only just met. For a new word you didn't know the first time, when its first review lands makes a big difference:

That is the forgetting curve at work: new words drop fastest in the first few days. Call one back the next day and about 60% are still there; wait a few more days and more than half have to be learned again.

Which means "how many days should I wait?" is the wrong question. The same three-day gap is too late for a difficult word you just met and far too early for one you already know. What should set the interval is the state of that individual card, not the calendar.

Three practical conclusions

1. Forgetting right after you learn something is normal. Forgetting is the default behaviour of memory, not a failure of studying. What decides the outcome is not whether you forget but whether you retrieve it before you do.

2. Weak memories need frequency; strong memories need distance. A new word left a week before its first review comes back less than half the time — that word needs to come back the next day. A word you already know is still 80–90% remembered after a week to a month, so reviewing it weekly is wasted effort. Move those spare repetitions onto the weak words and the total time stays the same while the result changes considerably.

3. Intervals should not be set in fixed days. A blanket "review every three days" over-reviews the easy words and under-reviews the hard ones. Both ends waste time.

One more point deserves its own line: re-reading is not reviewing. Seeing a word and feeling that you recognise it is a different state from being able to produce its meaning. Every number above measures active recall — try first, then check. That is also the only action that earns a longer next interval.

FAQ

What is the forgetting curve?

The forgetting curve describes how memory decays over time, and was proposed by Hermann Ebbinghaus in 1885. It shows that material you stop encountering is lost quickly over the first few days, then decays more slowly. It is the basis of spaced repetition: review before memory falls past the point of retrieval and you can hold onto it with far fewer repetitions.

Is the forgetting curve real?

Yes. In 2015, Murre and Dros at the University of Amsterdam published a replication of Ebbinghaus's experiment in PLOS ONE, and their curve came out very close to the 1885 one: forgetting is fast at first, then slows down. Our data shows the same shape: a new word you didn't know the first time is about 60% remembered if reviewed the next day, and less than half if the first review waits a week.

How much will I remember after a month?

It depends on whether that word was reviewed shortly before you would have forgotten it. Words reviewed on schedule are still 80–90% remembered after a week to a month; a new word whose first review waits a week comes back less than half the time.

Why do I still forget things I have reviewed?

Because a single review only pushes forgetting further out; it does not fix a memory permanently. A new word you didn't know the first time comes back only about 60% of the time, even when reviewed the very next day. What matters is that every review lands shortly before you would have forgotten, rather than long after you already have.

How does the forgetting curve relate to spaced repetition?

The forgetting curve is the phenomenon; spaced repetition is the response. Since memory decays at predictable points, schedule the review just before the decay reaches the threshold and you maintain the memory with the fewest repetitions. Algorithms like FSRS go further and estimate the decay rate for each card individually, so the timing tracks the moment you are actually about to forget.

How often should I review?

There is no single number of days that works for every word. Words you are still shaky on need to come back the next day; words you already know are still 80–90% remembered after a week to a month. The interval should follow each card's current memory strength rather than a uniform rule.

Is more review always better?

No. What works is one review shortly before you forget, not repetition while the memory is still sharp. Words you already know are still 80–90% remembered after a week to a month; reviewing inside that window buys very little memory for the time spent, and it crowds out the words that genuinely need the slot.

Does re-reading a word list count as reviewing?

It is far weaker. Recognising a word is a different memory state from producing its meaning. Every number in this article measures active recall — attempt first, then check the answer. Active recall is what earns a longer next interval; passive re-reading does not.

Can I apply the Ebbinghaus curve directly?

Not really. Ebbinghaus used nonsense syllables and no review at all, so what he measured is decay without intervention. Real vocabulary study involves words of differing strength that all get reviewed, so copying a fixed table of "20 minutes, 1 hour, 1 day" adds little. Use the shape of the curve as background; let the individual card state set the timing.

Source: Tomaru anonymous learning data (2026). Tomaru uses the FSRS algorithm to schedule every card. Further reading: What is active recall, What is the testing effect, What is FSRS.