The forgetting curve: what it is and how to flatten it

The forgetting curve is a graph of how fast you lose something you've learned. Hermann Ebbinghaus measured the first one in the 1880s, in experiments on his own memory. It drops steeply in the first hours and then levels off: a day after learning a list, he had to redo about two thirds of the work to learn it again, and a month later about four fifths. Testing yourself, spacing your reviews and sleeping well all make the curve flatter.

What Ebbinghaus measured

Ebbinghaus wanted to measure memory without the help of meaning, so he learned lists of nonsense syllables like WID and ZOF. He read each list aloud until he could recite it without a mistake and noted how long that took. Then he waited, from 20 minutes up to 31 days, and learned the same list again.

The second round was faster, because part of the list was still in his memory. The time he saved is what he measured. If learning a list took 10 minutes and relearning it a day later took about 6 minutes and 40 seconds, he had saved a third of the work. Psychologists call this the savings method.

What the numbers on the forgetting curve mean

Here is how much work Ebbinghaus saved when he relearned a list, from his own summary and a history of spaced repetition on SuperMemo's blog:

The table explains the shape of the curve. More than half of the forgetting happened in the first hour, and after the first day the losses were small.

These numbers measure relearning time, not how much he could recall. So when you read that people forget 70% of what they learn within a day, keep in mind that Ebbinghaus's 34% after one day means relearning took about two thirds of the original time. It doesn't mean he could recite a third of the list.

Is the forgetting curve real?

Yes. In 2015 Jaap Murre and Joeri Dros at the University of Amsterdam repeated the experiment. Dros, then 22, spent about 70 hours over 75 days learning and relearning 70 lists of nonsense syllables at the same intervals Ebbinghaus used. His curve came out similar to the original.

Murre and Dros also compared the curve with Ebbinghaus's data and earlier replications, and concluded that it most likely isn't smooth: it seems to jump up a little at about 24 hours, as if something between day zero and day one helped. They note that sleep research would predict this, without claiming they had proved it.

Both studies had a single subject learning meaningless material, so treat the exact numbers as Ebbinghaus's and Dros's, not yours.

The forgetting curve for things you read

Articles have meaning, which nonsense syllables don't, but the drop is still large. In a 2006 study at Washington University, students read a short prose passage and studied it again. Five minutes later they recalled 81% of it. A week later they recalled 40%.

The same study shows how to change the curve. Students who read the passage once and then tested themselves on it three times recalled 61% a week later.

The forgetting curve formula

Ebbinghaus fitted his results with a formula, also in chapter 7 of his book:

b = 100 × k ÷ ((log t)^c + k)

Here b is the work saved in percent, t is the time since learning in minutes, log is the base-10 logarithm, k = 1.84 and c = 1.25. For one day (t = 1,440 minutes) the formula gives about 30%, close to the 34% he measured. You'll also see the forgetting curve written as an exponential decay, R = e^(−t/S), where S stands for how stable the memory is. That version is a later simplification, and it's the idea behind spaced repetition schedulers: every successful review makes S larger, so the next drop is slower.

How to flatten the forgetting curve

Test yourself instead of rereading

Each time you pull an idea out of memory, it fades more slowly afterward. Close the article and write down what you remember, then check. The Washington University result above is one example, and active recall covers the method in detail.

Space out your reviews

One review right after reading helps, and later reviews help more. A 2006 review of 317 experiments by Nicholas Cepeda and colleagues found that reviews spread over time beat reviews crammed together. A simple schedule for something you read today is the next day, about a week later and about a month later. Reviewing at growing gaps like this is called spaced repetition, and the FSRS simulator shows how an app picks the gaps for each card. There's more on it in how to remember what you read.

Sleep after you learn

In NIH's article on sleep and memory, Harvard sleep researcher Robert Stickgold says that the sleep you get the night after you study is at least as important as the sleep you get the night before. If you read something important in the evening, go to bed, and do your first review the next day.

Questions about the forgetting curve

How much do we forget in 24 hours?

It depends on the material and on how well you learned it. In Ebbinghaus's experiments, relearning a list after 24 hours took about two thirds of the original time. In the Washington University study, students who had reread a prose passage lost about half of what they could recall at five minutes over the following week.

What is the 2-7-30 rule?

It's a review schedule: go over the material 2 days, 7 days and 30 days after you learn it. It doesn't come from one specific study, but it follows the spacing research and it's easy to remember.

How Bundle uses the forgetting curve

I made Bundle so that reviewing what you read takes a few minutes a day. When you finish an article, a Quick check asks you a few questions about it, and you keep the good ones as cards. Bundle schedules each card with FSRS, which estimates how likely you are to remember the card and brings it back when that chance gets low. Each time you remember a card, it waits longer before the next review.

Bundle is on the App Store for iPhone, and you can try it free for 14 days.