The Science of Spaced Repetition

You forget most of what you learn within a day. That's not a failure of discipline or intelligence — it's how human memory works. The brain aggressively discards information it doesn't expect to need again, and a single exposure is almost never enough to signal long-term importance.

Spaced repetition is the most researched and effective technique cognitive science has found for fighting this forgetting. It's not a study hack or a productivity trick — it's a method grounded in over 140 years of experimental research, validated across every subject and age group ever tested. This article covers that research: what we know about how memory works, how we got here, and how algorithms like FSRS turn decades of cognitive science into practical study schedules.

Ebbinghaus and the Forgetting Curve (1885)

The science of memory begins with one man memorizing nonsense. In 1885, Hermann Ebbinghaus — a psychologist at the University of Berlin — published Uber das Gedachtnis (Memory: A Contribution to Experimental Psychology), the first rigorous experimental study of human memory.

His method was deliberately tedious. Ebbinghaus memorized lists of nonsense syllables — meaningless consonant-vowel-consonant combinations like "DAX," "BUP," "ZOL" — to eliminate the effect of prior knowledge. He'd memorize a list to perfection, then test himself at increasing intervals: 20 minutes, 1 hour, 9 hours, 1 day, 2 days, 6 days, 31 days. He recorded how much he'd forgotten at each point, and how much effort it took to relearn the list.

The result was the forgetting curve: an exponential decay in retention. The drop is steepest immediately after learning — roughly 50-70% of new information is lost within the first 24 hours. After that, forgetting slows. What you remember after a week, you'll mostly still remember after a month.

But Ebbinghaus discovered something more important than the shape of forgetting: each review resets and flattens the curve. When he re-studied a list, the subsequent forgetting was slower. The second review made forgetting slower still. Each repetition deepened the memory trace, and the intervals between necessary reviews grew longer.

This is the foundational insight behind all spaced repetition: review timing is everything. Review too early, and you waste time reinforcing something you already remember. Review too late, and you're relearning from scratch — almost as much work as the first time. The optimal moment is right before you'd forget, when retrieval is still possible but requires effort. That effort is what strengthens the memory.

Why Spacing Works — Two Theories

Ebbinghaus showed that spacing works. It took another century of research to understand why. Two complementary theories dominate the literature.

Encoding variability

First proposed by William Estes in 1955 and developed by Arthur Glenberg in 1979, this theory argues that each time you encounter information, you encode it in a slightly different mental context. Your mood is different, the surrounding thoughts are different, even the physical environment may be different. Each review creates a new retrieval path to the same memory.

When you mass your practice — cramming the same material in a single session — every encoding happens in nearly identical context. You have one strong retrieval path that works in that specific situation but fails elsewhere. Spacing forces you to retrieve the memory across varied contexts, building a richer, more interconnected memory trace that's accessible from more starting points.

This is why typing an answer, hearing it spoken, and matching it in a quiz all strengthen the same memory differently — each mode creates a distinct encoding. Passive card flipping, by contrast, creates a single shallow trace.

Retrieval practice (the testing effect)

Henry Roediger and Jeffrey Karpicke published a landmark study in 2006 demonstrating that the act of retrieving a memory strengthens it more than re-reading does. In their experiment, students who studied a passage once and then took a recall test remembered significantly more one week later than students who studied the passage four times without testing.

The mechanism is straightforward: retrieval is effortful, and that effort triggers the brain to consolidate the memory more deeply. Re-reading feels productive — the material seems familiar — but it builds recognition, not recall. Recognition is knowing the answer when you see it. Recall is producing the answer from nothing. Spaced repetition forces recall, and every successful recall at a longer interval signals the brain that this information is worth keeping.

These two mechanisms work together. Spacing forces retrieval instead of recognition, and each retrieval in a new context builds a richer, more durable memory trace. This is why passive study modes — highlighting, re-reading, flipping cards without effort — consistently underperform active recall in every controlled study.

The Research — What 100+ Years of Studies Show

Individual studies can be compelling. Meta-analyses across hundreds of studies are conclusive.

Cepeda et al. (2006) published the most comprehensive meta-analysis of spacing effects in the Psychological Bulletin, synthesizing 254 studies spanning over a century. Their finding was unequivocal: distributing practice across time consistently improves retention compared to massing practice into a single session. This held across every subject tested (languages, math, science, motor skills), every age group (children through elderly adults), and every time frame (days to months). The optimal spacing gap increased with the desired retention interval — if you want to remember something for a year, space your reviews further apart than if you need it for a week.

Dunlosky et al. (2013) reviewed ten popular study techniques in Psychological Science in the Public Interest, rating each on effectiveness. The results were stark:

Distributed practice (spaced repetition) and practice testing (active recall) were the only two techniques rated "high utility." Highlighting and re-reading — the most popular study methods among students — were rated "low utility." The gap wasn't marginal; it was categorical.

Roediger & Karpicke (2006) demonstrated the testing effect directly. Students who took recall tests after studying retained 80% of material a week later, compared to 52% for students who spent the same total time re-reading. The students who re-read felt more confident in their knowledge. They were wrong.

These findings converge on a clear conclusion: spaced repetition combined with active recall isn't one technique among many — it's the most validated approach to memorization that psychology has produced. The research base spans over a century, thousands of participants, and every domain of learning.

From Index Cards to Algorithms

The science was settled by the mid-20th century. The challenge was turning it into a practical system.

Sebastian Leitner (1972) created the first systematic spacing method using physical boxes. Cards start in Box 1 (reviewed daily). Get a card right, and it moves to Box 2 (every 3 days), then Box 3 (weekly), and so on. Get it wrong at any stage, and it drops back to Box 1. Simple, intuitive, and a genuine improvement over random review — but the intervals are fixed and identical for every learner. A card that's easy for you gets the same schedule as one that's hard.

Piotr Wozniak and SM-2 (1987) brought computers into the equation. SM-2 gave each card its own "ease factor" — a multiplier that determined how fast intervals grew. Difficult cards got shorter intervals; easy cards got longer ones. This was the first personalized spacing algorithm, and it powered SuperMemo (later adapted by Anki) for decades. But SM-2's single-parameter model had a structural limitation: the ease factor controlled both difficulty estimation and interval growth, creating the "ease hell" problem where difficult cards got permanently stuck at short intervals. For the detailed technical comparison, see Why FSRS Is Better Than SM-2.

Jarrett Ye and FSRS (2022) represents the current state of the art. Instead of one parameter, FSRS tracks three — stability, difficulty, and retrievability — each independently adapted from your review history. The model was trained on millions of real review logs and peer-reviewed at ACM SIGKDD and IEEE TKDE. The result is more efficient scheduling: the algorithm predicts when you'll forget each card and times the review accordingly.

Each generation added more parameters and more data, producing more accurate predictions. The trend is clear: better models of memory lead to better scheduling, which leads to less wasted time and stronger retention.

How FSRS Turns Science into Scheduling

An algorithm's job is to answer one question for each card in your collection: when will you forget this? The more accurately it predicts the answer, the more efficiently it can schedule your reviews.

FSRS models a personalized forgetting curve for every card. Each card's stability determines how fast the curve decays — a well-learned card decays slowly, a new one decays fast. When your predicted recall probability drops to about 90%, a review is scheduled. This 90% target reflects what the research calls "desirable difficulty" — the point where retrieval is effortful enough to strengthen memory but not so late that you've forgotten entirely.

The visual below shows how this works in practice. Two cards learned on the same day reach the review threshold at different times because their stability differs — the algorithm gives each card exactly the interval it needs.

This is what Words on Repeat uses by default for all users. For technical details on how FSRS is implemented, see our spaced repetition guide.

What This Means for How You Study

The science points to specific, actionable principles:

Active recall beats passive review. Type the answer instead of flipping the card. If your app supports multiple study modes — typing, listening, matching — use them. Each mode engages different retrieval pathways, building the encoding variability that Glenberg's research predicts will improve retention.

Context strengthens encoding. Cards with example sentences, grammar notes, and usage context create richer memory traces than isolated word pairs. This is encoding variability applied at the card level — more context means more retrieval cues. Words on Repeat includes grammar notes and example sentences on every curated card for exactly this reason. For a deeper look at why learning from your own content produces even stronger memories, see Why Learning Words in Context Works Better.

Consistency beats intensity. Cepeda's meta-analysis showed that distributed practice outperforms massed practice at every duration tested. Ten minutes of daily review produces better long-term retention than a two-hour weekly cram session. The spacing itself is part of the mechanism — each day you return to the material is a new retrieval in a new context.

Trust the algorithm's scheduling. When FSRS says a card isn't due, reviewing it anyway provides minimal benefit. The algorithm is optimizing for the point where retrieval effort maximizes memory consolidation. Reviewing too early wastes time on cards you'd remember anyway; that time is better spent on cards that actually need attention.

The algorithm is not the whole picture. Spaced repetition handles memorization — the "what" of a language. Comprehension, conversation practice, and immersion handle the "how" and "why." The most effective learners use SRS to build and maintain their vocabulary foundation, then spend the rest of their study time on skills that flashcards can't teach. For a practical look at how to structure your study sessions, see our complete guide to learning vocabulary with Words on Repeat.

Frequently Asked Questions

References

1. Ebbinghaus, H. (1885). Memory: A Contribution to Experimental Psychology. Translated by H.A. Ruger & C.E. Bussenius, 1913. psychclassics.yorku.ca
2. Roediger, H.L. & Karpicke, J.D. (2006). "Test-Enhanced Learning: Taking Memory Tests Improves Long-Term Retention." Psychological Science, 17(3), 249–255. doi:10.1111/j.1467-9280.2006.01693.x
3. Cepeda, N.J., Pashler, H., Vul, E., Wixted, J.T. & Rohrer, D. (2006). "Distributed Practice in Verbal Recall Tasks: A Review and Quantitative Synthesis." Psychological Bulletin, 132(3), 354–380. doi:10.1037/0033-2909.132.3.354
4. Dunlosky, J., Rawson, K.A., Marsh, E.J., Nathan, M.J. & Willingham, D.T. (2013). "Improving Students' Learning With Effective Learning Techniques." Psychological Science in the Public Interest, 14(1), 4–58. doi:10.1177/1529100612453266
5. Settles, B. & Meeder, B. (2016). "A Trainable Spaced Repetition Model for Language Learning." Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (ACL). aclanthology.org/P16-1174
6. Ye, J. (2022). "A Stochastic Shortest Path Algorithm for Optimizing Spaced Repetition Scheduling." ACM SIGKDD Workshop. doi:10.1145/3534678.3539081 — See also Why FSRS Is Better Than SM-2 for the detailed algorithm comparison.