The Forgetting Curve Explained: What Ebbinghaus Found, What It Means, and How to Flatten It

The forgetting curve shows that memory for something you just learned falls off quickly at first and then much more slowly. Hermann Ebbinghaus mapped it in 1885: within an hour he had lost a large share of a freshly learned list, but what survived the first day faded only gradually over the following month. The practical lesson is simple. Forgetting is fastest right after learning, so the first reviews matter most, and every successful review slows the next round of forgetting.

Who was Ebbinghaus and what did he actually do?

Hermann Ebbinghaus was a German psychologist who wanted to study memory with the same rigor as physics. His book Über das Gedächtnis (published in English as Memory: A Contribution to Experimental Psychology) reported years of experiments in which he was the only participant.

To strip out the effect of meaning and prior knowledge, he invented nonsense syllables: consonant-vowel-consonant strings such as DAX, BOK or ZEF. He learned lists of them by reading them aloud at a steady pace until he could recite the whole list twice without error. Then, after a set delay, he relearned the same list and recorded how long it took.

The key measure was savings: the percentage of the original learning time he no longer needed. If a list took 1,000 seconds to learn and 600 seconds to relearn, savings was 40%. This is subtle and important. Savings can be positive even when you cannot consciously recall a single item, because some trace of the memory still makes relearning faster.

The numbers behind the curve

Ebbinghaus tested seven delays. His savings scores, as they are usually reported, looked like this:

Two features stand out. First, the loss is front-loaded: most of what disappears over a month is already gone within the first day. Second, the curve does not reach zero. A month later, relearning was still about a fifth faster than learning from scratch. Ebbinghaus fitted a logarithmic function to these points; later researchers have argued for power functions and other shapes, but all of them share this "steep, then shallow" profile.

You can experiment with the shape yourself in our forgetting curve calculator, which shows how retention changes with memory strength and the number of reviews.

Did the forgetting curve replicate?

Many psychology findings from the early era have not survived modern scrutiny. This one has. In 2015, Jaap Murre and Joeri Dros published a direct replication in PLOS ONE. One of the authors spent around 70 hours learning and relearning lists of nonsense syllables using Ebbinghaus's procedure and the same delays.

The resulting curve was strikingly similar to the 1885 original. The replication also confirmed something Ebbinghaus had seen in his own data: a small upward bump at the 24-hour point, where retention was slightly better than the smooth curve predicts. Murre and Dros suggested that sleep between learning and testing may explain it, which fits a large body of later research on memory consolidation (more on this in our post on sleep and vocabulary memory).

Beyond direct replications, the general shape appears again and again in studies of word pairs, facts, faces and real-world knowledge. Harry Bahrick's long-term studies of school Spanish, for example, found steep losses in the first few years after a course ended, followed by a stable core of vocabulary that people still knew decades later.

What the forgetting curve does not mean

Because the curve is easy to draw, it gets attached to claims it cannot support. Four are worth correcting.

"You forget 70% of everything within 24 hours"

This figure circulates widely in training and marketing material. It comes from treating Ebbinghaus's savings scores as if they were recall percentages and then applying them to all learning. His data describe relearning time for meaningless syllables learned by rote. They are not a universal constant for lectures, conversations or vocabulary.

"Meaningful material follows the same curve"

Ebbinghaus chose nonsense syllables precisely because they have no meaning to hook onto. Real words, especially ones met in a memorable sentence or linked to things you already know, usually fade more slowly. Ebbinghaus himself noted that memorizing meaningful poetry took far less effort than nonsense lists.

"There is one curve for everyone"

The speed of forgetting depends on how well something was learned, how it was encoded, its similarity to other memories (interference), sleep, and how memory is tested. Recognition ("Have you seen this word?") stays high much longer than free recall ("What is the word for this?"). There is a family of curves, not one.

"Forgotten means gone"

The savings method exists because forgotten material often leaves a trace. A word you cannot recall today may still be relearned in seconds. This is good news for learners: a lapse is not a reset to zero.

Why the curve matters for vocabulary

New words are close to the worst case for memory. A word like ephemeral is an arbitrary link between a sound, a spelling and a meaning, with few existing hooks unless you know a related word. If you look it up once and move on, you will usually lose access to it within days unless you happen to meet it again.

That is why so many learners feel they "look up the same word five times". They are living on the steep part of the curve. The fix is not to try harder during the first encounter, but to arrange a second and third encounter before the memory drops too far.

How reviews flatten the curve

Ebbinghaus did not only describe forgetting. He also showed that repeated learning sessions spread over several days produced more durable retention than the same amount of practice crammed into one sitting. Later research has turned this into two principles that work together.

1. Retrieve, do not reread

Trying to recall a word's meaning from memory strengthens it far more than looking at the definition again. This is the testing effect, covered in depth in our article on the testing effect. A review that requires effort, and ideally succeeds, is the one that changes the curve.

2. Space reviews at expanding intervals

After each successful retrieval, the memory decays more slowly, so the next review can wait longer. This is the spacing effect, and it is the foundation of spaced repetition. A typical expanding schedule for a new word looks like this:

If a review fails, shorten the interval and build back up. Algorithms such as SM-2 and FSRS automate this per word, and our spaced repetition scheduler lets you generate a plan by hand. For the full research story behind interval choice, see the spacing effect research.

Natural encounters count as reviews too

A review does not have to be a flashcard. Meeting ephemeral again in a news article, noticing it and remembering what it means, is a retrieval event in context, and it may be more valuable than a card because it also shows you how the word is used. The difficulty is that mid-frequency words come back irregularly, sometimes not for months, which is long enough for the curve to win.

A practical approach combines both: deliberate recall to catch words during the steep early period, and plenty of reading so words keep reappearing in context. Wordie supports the second half of this by highlighting words you have saved whenever they show up again on a web page, so a chance encounter becomes a noticed one, and its recall mode lets you test yourself on the words you captured.

A simple plan based on the curve

1. Capture the word with its sentence the first time you meet it, so you have a meaningful context to remember.
2. Test yourself within 24 hours. Look at the word, try to recall the meaning, then check.
3. Expand the gaps after each success: roughly 3 days, 1 week, 2–3 weeks, then monthly.
4. Shrink the gap after a miss and do not count the lapse as failure; relearning is faster thanks to savings.
5. Read widely so the word reappears naturally and you see it in new contexts.

For more everyday tactics, see how to stop forgetting words.

The bottom line

The forgetting curve is real, robust and well replicated: memory for new material drops fast, then slowly, and some trace usually remains. It is also widely misquoted. It does not prove that everyone forgets a fixed percentage in a day, and it describes meaningless syllables learned by rote rather than words learned in context. The useful message is that forgetting is predictable, and therefore manageable. Retrieve new words early, space later reviews further apart, and let reading supply extra encounters, and the curve flattens enough that a few minutes of review keeps a word for years.

Sources: Ebbinghaus, H. (1885), Über das Gedächtnis (English translation 1913, Memory: A Contribution to Experimental Psychology); Murre, J.M.J. & Dros, J. (2015), "Replication and analysis of Ebbinghaus' forgetting curve", PLOS ONE; Bahrick, H.P. (1984), "Semantic memory content in permastore: Fifty years of memory for Spanish learned in school", Journal of Experimental Psychology: General; Roediger, H.L. & Karpicke, J.D. (2006), "Test-enhanced learning", Psychological Science; Cepeda, N.J. et al. (2006), "Distributed practice in verbal recall tasks", Psychological Bulletin.