How many times should you practice one thing?

“Practice more” is advice nobody needs. The useful question is how much: how many times should you answer, recall, or apply one idea before your time is better spent elsewhere? For a single study session, memory research gives an unusually concrete answer. It is smaller than most people fear, and larger than most people practice.

Three successful retrievals, then stop

For one sitting, the best-supported target is about three successful retrievals per idea: you produce the answer correctly, from memory, three times, with other work in between. Stopping after one success leaves the memory fragile. Pushing past three buys almost nothing. Durability after that comes from re-checking on later days, not from more repetitions today.

Where the number comes from

In Vaughn and Rawson’s experiments, published in Psychological Science, learners practiced vocabulary until they had retrieved each item correctly one, three, five, seven, nine, or eleven times. Later memory climbed steeply from one success to three, then flattened. Past three, extra correct retrievals in the same session added close to nothing, with one exception worth knowing: unusually difficult material can justify more.

The floor matters as much as the ceiling. In Karpicke and Roediger’s 2008 study, items dropped from practice after a single correct answer were later recalled less than half as often as items that kept being tested. One success feels like knowing. It is not yet remembering.

Count successes, not attempts

The unit is a correct retrieval, not an exposure. Rereading does not count: in the same 2008 study, extra restudying of already-known items had no measurable effect on later recall. A failed attempt does not count toward the three either, but it is not wasted, because an error followed by feedback and a corrected attempt is one of the better-documented ways to learn.

What matters after a miss is what happens next. Reading the identical explanation again mostly reproduces the identical confusion. The pattern with better support: see why the answer was wrong, get the idea explained differently than the first time, then retry at a level where you can succeed, and build back up from there.

Today’s three are not forever

Three successes today does not make a memory permanent. The most durable schedule researchers have tested, called successive relearning, pairs an initial session like the one above with short re-checks on later days, each ending once you retrieve the idea correctly again. Relearning is fast. An idea that took three attempts to earn today usually takes one to re-earn on Thursday, and the spacing is what converts a session’s work into something that lasts. It will not feel that way while you do it; why the feeling misleads is its own story.

What this means in practice

• Within one sitting, aim for about three correct productions of each new idea, separated by other work rather than back to back.
• Vary the form. Recalling a definition, applying it to a case, and explaining it in your own words are three different successes, and the harder forms tell you more.
• After three, stop. A fourth repetition is worth less than the same minutes spent on a different idea, or on a re-check two days from now.
• If you use flashcards, resist deleting a card the moment you get it right, and resist grinding one card ten times in a row. Both instincts are wrong, in opposite directions.

Where Ulern stands

We build Ulern, a learning system, so read this section knowing that.

Ulern’s teaching loop is built to land on that number. Something new is taught once, then confirmed with an easy first activity. Later in the session it returns for the check that counts, where you produce the answer at the level you set out to reach. That is about three activities per idea, and in our internal test runs the average comes out at three almost exactly. A miss adds roughly two more steps, following the recovery pattern above: feedback, a fresh explanation from a different angle, then an easier retry. The same explanation is never repeated, and a miss is never followed by a harder question. We have not yet run an external trial; the studies above are why the loop has the shape it has.