Active Recall: Why Retrieval Works Better Than Rereading

Imagine two people learning the same chapter. One reads it four times. The other reads it once and then tries three times to write everything down from memory. Who knows more after a week?

If you answered "the second," you're right. If you wonder why almost everyone chooses option one anyway, you're right about that too. This article explains both.

What Active Recall Is

In research, active recall is called retrieval practice: you pull previously learned material out of memory. The learning advantage that results later is called the testing effect. The typical experimental design is simple: learn first, then either retrieve or study again, followed by a final test after a delay.

"Retrieving" does not mean "writing an exam." A flashcard you flip over after formulating the answer in your head is retrieval practice. A blank sheet on which you write what you still remember from the chapter counts too. A lecture handout you go over again with a highlighter does not.

The Key Experiment: Who Laughs Last

Roediger and Karpicke tested this in 2006 with students. In Experiment 1, 120 students read two expository texts. They reread one and tried to recall the other freely, without feedback. Testing took place after five minutes, two days, or one week.

Final test after	Rereading	Retrieval
five minutes	81 %	75 %
two days	54 %	68 %
one week	42 %	56 %

After five minutes, rereading won. After two days and after one week, retrieval came out ahead, with effect sizes of d = 0.95 and d = 0.83. In Experiment 2 with 180 people, three retrievals beat four readings after one week, with d = 1.26.

Here lies the trap. Rereading feels good and is genuinely better in the short term. Anyone who checks in the evening, right after studying, whether something "sticks" gets rewarded for rereading. The bill arrives later.

To be fair, the study had limits: two short texts, observation over one week at most, and in Experiment 2 the retrieval phases took longer than the study phases, so total study time was not exactly equal.

Why Everyone Rereads Anyway

Karpicke, Butler, and Roediger asked in 2009: 177 students described their learning strategies. 84% named rereading, and for 55% it was the most common strategy. Only 11% clearly described retrieval practice. This is a survey based on self-reports, but the picture is clear.

More interesting is an experiment by Karpicke from the same year. Learners could decide for themselves whether to keep studying a vocabulary pair, keep testing it, or remove it from the stack. The result: many stopped too early, after the first successful retrieval. Those who kept retrieving had 78% correct after one week. Those who kept studying had 39%. Those who removed the card had 35%.

So a successful retrieval does not mean "I can do it." It means "I can do it right now."

Getting It Right Once Is Not Enough

Karpicke and Roediger sharpened this point in Science in 2008. 40 students learned 40 Swahili-English word pairs. Once a word had been recalled correctly, it was either kept in study and testing, tested only, studied only, or removed entirely.

After one week, the two conditions with continued retrieval reached roughly 80%. Without further tests, the figures were 36% and 33%. Further study of words already known brought no measurable advantage under these conditions.

What you should not conclude from this is "rereading does nothing at all." The study examined rereading after successful retrieval, with vocabulary and a small sample. What you can conclude: don't stop quizzing a card just because you knew it once. That is exactly the job of a spaced repetition algorithm. How it decides when a card comes back is explained in the guide Spaced repetition explained.

How Large Is the Effect Really?

Individual experiments are nice, but meta-analyses are more honest. The key question is always: retrieval compared to what?

• Against restudying: Rowland (2014) found, across 159 effect sizes from 61 studies, an average Hedges' g of 0.50. The studies differed widely, so the average does not represent one uniform effect.
• Against broader control conditions: Adesope et al. (2017), with 118 articles and 15,427 participants, arrived at g = 0.61 overall. Against rereading it was g = 0.51, against doing nothing g = 0.93. The often-cited 0.61 is therefore not a pure comparison with rereading.
• In real classrooms: Yang et al. (2021) analyzed 222 studies with 48,478 learners from classroom contexts: g = 0.499 overall, against restudying g = 0.330.
• Against other active learning methods: Here the evidence becomes thin. Against elaborative study strategies, the advantage in Yang et al. was only g = 0.095 and not statistically significant.

The defensible conclusion is: on average, retrieval improves later retention compared with rereading. "Active recall is superior to every other method" goes beyond the data.

Retrieval Versus Mind Maps

A popular argument for active recall is Karpicke and Blunt (2011). Retrieval practice outperformed concept mapping (mind maps made with the text in view) with d = 1.50. In Experiment 2, 101 of 120 people performed better after retrieval practice, even on a test that itself required a concept map.

The story has a second chapter, though. In 2023, Mayrhofer et al. ran a preregistered study with 230 analyzable participants and matched study time and instructions. On text-близ questions, retrieval practice scored 73% and concept mapping 70%, with no significant difference. This does not prove the two are equally good. But it weakens the blanket claim that retrieval is inherently superior to mind maps.

Does This Also Help With New Questions?

The reasonable worry: if I study this way, am I only learning to answer my own flashcards? Pan and Rickard (2018) examined this in a meta-analysis with 10,382 participants. Result: retrieval also improves performance on material that goes beyond the original practice question, d = 0.40. For application and inference questions it was d = 0.32, with large differences between studies.

The limit: some types of transfer were barely studied. The analysis does not show far transfer to arbitrary new problems. If you need to be able to apply knowledge, practice application. So also write cards that ask about the why and the how, not only the what. Examples can be found in the guide How to write good flashcards.

Feedback: The Underrated Part

Retrieving without checking the answer is only half the method. In Rowland's analysis, the advantage with feedback was g = 0.73, without feedback g = 0.39. Without feedback, and when at most half of the answers were initially correct, it was practically zero (g = 0.03). These are patterns from comparing many studies, not a guaranteed formula, but the direction is clear: if you recall something wrongly and don't notice, you learn the error.

Butler and Roediger (2008) demonstrated this with multiple-choice tests. Without feedback, learners sometimes later adopted wrong answer options. With feedback, whether immediate or delayed until the end of the test, this dropped to the level of untested questions. Feedback delivered at the end of the practice test actually did slightly better than immediate feedback (d = 0.56). A general recommendation to postpone feedback as long as possible does not follow from this.

Guessing Helps When the Answer Follows

It sounds paradoxical, but it is supported: Kornell, Hays, and Bjork (2009) had learners guess answers before seeing the correct one. In Experiment 5, with 30 people and equal study time, after an average of 38 hours they recalled 47% correctly after a guessing attempt, compared with 35% after direct study (d = 0.94).

The limits matter: small samples, simple word pairs, and not every experiment in the series showed the advantage. The careful formulation is: an unsuccessful attempt can enhance subsequent learning when the correct answer follows. Uncorrected errors do not help.

For you, this means: a new card you don't know is not a lost round. Still try to give an answer before flipping.

Open Question or Multiple Choice?

A common claim is that free recall always works better than recognition. Rowland did find differences in a subanalysis: free recall g = 0.81, recall with cues g = 0.72, recognition g = 0.36. In classrooms, Yang et al., by contrast, found practically identical advantages for recall (g = 0.520) and recognition (g = 0.518).

So there is no universal winner among formats. With multiple choice, though, you need feedback so that wrong options don't stick.

How to Use Active Recall Today

1. Formulate the answer, then flip. In your head or aloud, but concretely.
2. Judge honestly. "Almost knew it" is not knowing it.
3. Don't stop after the first hit. A card you knew once must come back.
4. Guessing is allowed. For new cards, one attempt first, then the solution.
5. Add why-questions. Fact cards alone are not enough for transfer.
6. Blank sheet. After each chapter, write down what you still know, then compare with your lecture handout.

Active Recall with checkst

checkst is built around exactly this loop: see the question, retrieve the answer, flip, judge honestly. FSRS then plans when the card returns, so you don't stop after the first hit. If writing cards takes too long, you can have the AI suggest cards from your lecture handout, a PDF, or a photo, and go through them one by one before saving. And if you study with a partner, you can share a deck in a study group. The retrieval itself remains your job, though, and that is the part that works.

For the bigger picture, of how retrieval, spacing, interleaving, and sleep work together, see the guide How should you learn best?.