Spaced repetition: more than a memorisation technique
Spaced repetition is a learning method that schedules reviews at increasing intervals, calibrated to how well you know each piece of information. Instead of reviewing everything on a fixed calendar, the algorithm adapts: easy items come back after weeks or months; difficult ones after a day or two. The result is a system that invests review time exactly where it is most needed.
From Ebbinghaus to FSRS: a 140-year history
The scientific foundation dates to 1885, when Hermann Ebbinghaus published the first systematic study of memory and forgetting. Working alone, using himself as the sole subject, he memorised thousands of nonsense syllables and measured how quickly he forgot them. The result -- the forgetting curve -- showed that memory decays exponentially, with the steepest drop occurring within the first 24 hours. He also discovered the spacing effect: distributing review sessions over time rather than massing them together produced dramatically better retention for the same total number of repetitions.
This observation remained largely theoretical for nearly a century. In 1972, Sebastian Leitner turned it into a practical system using numbered card boxes -- cards answered correctly moved to a higher box reviewed less frequently; cards answered incorrectly dropped back. In 1987, Piotr Wozniak wrote the first computerised SRS algorithm (SM-2), which became the basis for Anki and dozens of other applications. Modern algorithms like FSRS, developed through machine learning on millions of real review sessions, now model individual forgetting curves with far greater precision than any fixed schedule could achieve.
The core mechanism: compounding intervals
The mechanism is deceptively simple: review each item just before you would forget it. If retrieval succeeds, the next interval expands -- from one day, to three days, to a week, to a month, eventually to years. If retrieval fails or is difficult, the interval resets to something short and the card is treated as partially relearned.
What makes this compounding dynamic powerful is its long-term trajectory. Early reviews feel slow: you are reviewing things you still remember quite well. But after 6 to 8 well-timed reviews, a memory that once decayed within 24 hours can remain accessible for months or years with no additional attention. The total time invested is a fraction of what repeated cramming would require -- and the retention at the end is incomparably stronger.
Cepeda et al. (2006) analysed 317 experiments with 839 effect-size estimates on distributed versus massed practice: distributed practice improved retention in 96% of studies. Under most tested conditions, spaced practice outperforms massed practice by a factor of 2 to 3 at a 30-day retention test. Few findings in cognitive psychology are this consistent across age groups, domains, and methods.
Cepeda et al. (2006). Distributed Practice in Verbal Recall Tasks. Psychological Bulletin, 132(3), 354-380.The forgetting curve: why everything fades so fast
Without review, the brain discards information rapidly. Ebbinghaus found that within 20 minutes of learning, roughly 40% is already gone. Within a day, that rises to 70%. Within a week, over 77%. This pattern -- steep initial drop, then gradual levelling -- holds across ages, subjects, and individuals, though exact rates vary with content type and prior knowledge.
- After 20 minutes: ~40% forgotten
- After 1 hour: ~56% forgotten
- After 1 day: ~70% forgotten
- After 1 week: ~77% forgotten
- After 1 month: ~79% forgotten
Why the brain forgets: an adaptive mechanism
Forgetting is not a malfunction -- it is the brain's efficiency mechanism. Memory storage is metabolically costly, and the brain actively prunes what it deems irrelevant based on how often information is accessed. A memory that is never retrieved is interpreted as unneeded; synaptic connections weaken through disuse until the trace becomes inaccessible. This is adaptive: the brain evolved to retain survival-relevant information, not exam material.
This is why passive exposure -- reading a textbook, attending a lecture, watching a video -- produces such limited long-term retention without follow-up. A single activation of each piece of information is enough for the brain to classify it as low-priority. Spaced repetition sends the opposite signal: each review says this information is needed repeatedly, triggering progressively more durable consolidation.
The savings effect: why relearning is faster than first learning
Ebbinghaus measured not only the forgetting curve but what he called the savings effect: even when a memory appears entirely forgotten, residual traces remain that make relearning significantly faster than original acquisition. A word set that took 12 repetitions to first learn might take only 6 repetitions to relearn perfectly after a period of forgetting -- even if the learner cannot consciously recall any of the words.
This has a practical implication for spaced repetition: early reviews, even ones that feel unnecessary because the memory still seems fresh, build savings that make all future reviews more efficient. Skipping early reviews does not merely delay repetitions -- it erases traces prematurely, turning later sessions back into initial learning rather than maintenance.
Each spaced review after successful retrieval roughly doubles the time before the next review is needed. This compounding effect is why learners who use spaced repetition consistently outperform those who cram: the total number of reviews required to maintain a stable memory decreases as the interval grows, producing better retention with less study time over a long horizon.
Why rereading creates an illusion of mastery
The most common obstacle to adopting spaced repetition is not laziness -- it is a cognitive bias: rereading creates a false sense of mastery that feels indistinguishable from genuine learning, and feels much more comfortable than active retrieval.
Perceptual fluency: why smooth reading feels like knowing
When you reread a passage, the words flow easily. You recognise sentences, you tell yourself 'yes, I know this,' and the ease of processing creates an impression of competence. Psychologists call this perceptual fluency: the smooth, effort-free experience of rereading is mistaken for a signal that the material is well-learned. It is not.
Fluency is a property of the reading experience, not of memory strength. Rereading the same page five times makes it feel extremely familiar -- but familiarity is recognition, not retrieval. On an exam, in a conversation, or when you need to explain something without notes, the question is never whether you recognise the information. It is whether you can produce it from scratch, with no external cues.
Recognition vs retrieval: a difference that matters
Recognition and retrieval are fundamentally different cognitive processes. Recognition is triggered by seeing the answer -- multiple-choice questions where you select from options, or rereading where the information is in front of you. Retrieval requires generating the answer with no external cue, navigating your own memory networks to reconstruct it from scratch.
Spaced repetition, by design, puts you in retrieval mode every single time: the question appears, the answer is hidden, and you must produce it before checking. This is uncomfortable, especially early in learning. That discomfort is precisely the point. Research consistently shows that the more effortful the retrieval -- without being impossible -- the stronger the resulting memory consolidation. Bjork calls this desirable difficulty: harder study, when the difficulty is of the right kind, produces better long-term retention.
Rereading feels productive because it is easy, fast, and comfortable. It produces a genuine signal of recognition that is nearly impossible to distinguish from retrieval ability -- until the moment you need to recall something without the text in front of you. The test: can you recall the answer before seeing it? If looking at the answer surprises you, you were testing retrieval. If you already knew what would appear, you were practising recognition.
What spaced repetition changes in the brain
Spaced repetition works because it aligns with two well-documented neurobiological mechanisms: the testing effect and sleep-based memory consolidation. Understanding them explains not just why spacing works, but why the specific pattern of spaced retrieval matters.
The testing effect: retrieval is consolidation
The testing effect -- also called the retrieval practice effect -- is the finding that retrieving information from memory is a more powerful consolidation event than re-exposing yourself to the same information. Roediger and Karpicke (2006) showed that students who took a practice test after reading a passage retained 65% of the material one week later, versus 40% for students who reread the passage the same number of times. The gap widens further at one month.
The mechanism is neurobiological: when you retrieve information, the neural pathways associated with that information are activated and strengthened via long-term potentiation (LTP). Each successful retrieval makes the next retrieval faster and more reliable. The act of retrieval is not a measurement of learning -- it is the primary driver of learning. Spaced repetition is simply a system for creating repeated, well-timed retrieval opportunities at maximum efficiency.
Sleep consolidation: seven cycles per week
Memory consolidation -- the process by which recently acquired information becomes stable long-term memory -- occurs primarily during sleep. During slow-wave sleep, the hippocampus replays the neural activation patterns from the day's learning sessions, progressively transferring memories to cortical long-term storage. During REM sleep, memories are integrated into existing knowledge networks and creative connections form.
This means that consistent daily sessions have a structural advantage over occasional long sessions: each daily session provides material for one overnight consolidation cycle. A learner who reviews for 15 minutes each evening gives their brain seven consolidation cycles per week. A learner who reviews for 90 minutes once a week gets only one. Frequency matters more than duration -- which is exactly what spaced repetition's design optimises for.
Stickgold (2005) showed that sleep is not passive rest but an active memory-processing state. Participants who slept between a learning session and a retention test showed 20-40% better recall than those who stayed awake, regardless of total elapsed time. Combining retrieval practice with adequate sleep produces effects that neither mechanism alone can match.
Stickgold, R. (2005). Sleep-dependent memory consolidation. Nature, 437, 1272-1278.SM-2 and FSRS: how algorithms calculate your reviews
Not all spaced repetition algorithms are equal. Two dominate modern flashcard software: SM-2, the algorithm that defined the field for three decades, and FSRS, a newer approach trained on real review data that models forgetting curves with greater accuracy.
SM-2: the algorithm that built the field
SM-2 (SuperMemo-2) was developed by Piotr Wozniak in 1987 and made public in 1990. It introduced the ease factor -- a per-card multiplier that starts at 2.5 and adjusts based on performance. After a correct answer, the next interval is multiplied by the ease factor. After a difficult answer, the ease factor decreases and the interval resets. This simple model allowed Anki and dozens of other applications to provide automated scheduling for the first time.
SM-2 has been validated over more than three decades of real-world use and works reliably for most learners and content types. Its known limitations: it does not independently model memory stability and retrievability, it can over-schedule easy cards and under-schedule difficult ones, and its ease factor adjustments can create a situation where cards rated difficult get trapped in short-interval loops that are hard to escape.
FSRS: machine learning meets the forgetting curve
FSRS (Free Spaced Repetition Scheduler) was developed by Jarrett Ye starting in 2022, trained on tens of millions of real review sessions from Anki users. It independently models two properties of each memory: stability (how long it will persist before significant decay) and difficulty (the card's inherent resistance to consolidation). These parameters are estimated from each learner's actual review history and updated after every session.
In practice, FSRS schedules reviews at the moment when predicted retention drops to a configurable threshold (default 90%). This produces fewer over-reviews of well-known cards and more timely reviews of difficult ones. Comparative analyses show FSRS achieves better retention per unit of review time than SM-2, particularly for learners with larger decks where interval calibration matters most. Memia uses FSRS.
Both SM-2 and FSRS are dramatically more effective than reviewing without any algorithm -- and both are dramatically more effective than not reviewing at all. The choice between them matters at the margin; the bigger gains come from doing spaced repetition consistently, with well-designed cards, over months rather than weeks.
What spaced repetition is -- and is not -- good for
Spaced repetition is highly effective for specific types of content and learning goals. Understanding its scope prevents frustration when applying it to the wrong domain and helps direct it where the returns are highest.
Language learning: the strongest evidence base
Vocabulary acquisition is arguably the most studied and best-supported application of spaced repetition. Learning a language requires mastering thousands of word-meaning pairs -- exactly the kind of declarative, one-to-one knowledge that SRS handles best. Research on vocabulary learning consistently shows spaced repetition produces retention rates 2 to 3 times better than traditional study methods at a one-month horizon, with the gap widening as the time horizon extends.
The key variable in language learning is card direction: cards should test production (seeing an English prompt, producing the target-language word) rather than only recognition. Listening comprehension and pronunciation require different types of practice -- but the vocabulary foundation that supports both benefits enormously from a consistent SRS routine, making every hour of immersion or conversation practice more efficient.
Knowledge-heavy exams: start early, build incrementally
Knowledge-heavy exams -- medical licensing, bar exams, professional certifications, technical credentials -- are ideal territory for spaced repetition. The volume of material is large, the time horizon is measured in weeks or months, and the content is highly factual (definitions, dosages, statutes, protocols). All three factors play to SRS's strengths.
The critical variable is starting early. A learner who begins 8 weeks before an exam and adds 20 cards per day builds a deck of 1,120 cards by exam day, each reviewed multiple times at optimal intervals. A learner who starts 2 weeks before and adds 80 cards per day reaches the same deck size -- but most cards have been reviewed only once or twice, with insufficient spacing to produce durable retention. Same total volume, vastly different outcome.
What spaced repetition does not replace
Procedural knowledge -- debugging code, performing a surgical technique, writing a persuasive argument, playing an instrument -- requires deliberate performance practice with feedback, not flashcard review. Active recall strengthens the declarative layer of procedural skills (knowing what to do and why), but does not replace actually doing it under realistic conditions.
Analytical reasoning, critical thinking, and synthesis also fall outside SRS's effective range. Understanding why a historical event occurred, evaluating competing economic theories, or solving novel mathematical problems require working through problems and building flexible mental models -- not memorising fixed question-answer pairs. Use spaced repetition for the knowledge foundation; use problem-solving practice for the reasoning layer above it.
How to apply spaced repetition: building a routine that sticks
A realistic approach starts small. The goal is not to create 300 cards on the first night -- it is to build a lightweight system you can maintain consistently for months.
The most common failure is adding too many new cards too fast. This creates a compounding queue: sessions become overwhelming within two weeks, and learners abandon the system entirely.
- Create a focused deck: one topic, 30-50 cards to start. Use Memia to generate cards from any content.
- Each card should have one clear question and one clear answer. Avoid cramming multiple facts into one card.
- Do your due reviews every day -- even 5 minutes counts. Consistency beats intensity.
- Add new cards only when your daily queue feels manageable (under 15 minutes).
- Review card quality over time: split vague cards, clarify ambiguous questions, add examples.
Card design: one question, one answer
Good cards follow a single rule: one question, one answer. Avoid cramming multiple facts into a single card -- the algorithm cannot distinguish which part you remembered and which you missed, and compound cards build fragile memories that collapse when context changes. Split compound cards immediately. A card asking for three causes of the French Revolution should become three separate cards.
The question should match how you will actually need the information. If you need to produce a French word from its English meaning in conversation, test production (English prompt, French answer) rather than only recognition (showing both and confirming they match). The format of the question determines the type of memory built -- and only memories that match real-world retrieval conditions transfer reliably to performance.
Building the habit: minimum viable consistency
Consistency beats intensity. A 10-minute review session every day outperforms a 70-minute session once a week -- not because of total time, but because of seven overnight consolidation cycles versus one. Build the review into a fixed slot: morning coffee, commute, or right before sleep. The goal is to make it as automatic as brushing your teeth.
The minimum viable session is clearing whatever the algorithm marks as due. Some days that is 5 minutes; others 20, depending on how many cards come due simultaneously. The only firm rule: clear the due queue before adding new cards. New cards expand future sessions; due cards contain memories at their optimal review moment that, if missed, have to relearn rather than consolidate.
The 4 mistakes that make spaced repetition fail
Spaced repetition fails for predictable reasons. Each of these four mistakes is correctable once you know what to look for.
Adding too many cards too fast
This is the most common failure mode. Adding 20 to 30 new cards per day creates a compounding review queue: two weeks later, all those cards come due simultaneously, sessions exceed 30 minutes, and most learners abandon the system entirely. The backlog feels insurmountable; the tool that was supposed to make learning easier becomes another source of stress.
The fix is a hard limit on daily new cards -- 5 to 10 is a safe starting point. Raise this limit only when your review queue consistently takes under 15 minutes to clear. Growth should feel gradual and sustainable. A smaller active deck reviewed consistently produces better retention than a large deck that overwhelms the routine.
Poor card design
Cards that are too long, contain multiple facts, or ask ambiguous questions are hard to review and give a false sense of progress. A card you can answer by pattern-matching a specific phrase without understanding the underlying concept builds recognition, not retrieval -- and will fail precisely when it matters most.
Audit your deck periodically. Split any card covering more than one distinct fact. Rewrite questions you answer by pattern rather than recall. Add concrete examples to abstract definitions. A smaller deck of well-designed cards outperforms a large deck of mediocre ones by every measure: review speed, retention rate, and transfer to real-world use.
Skipping sessions
Missing sessions is more costly than it appears. Skipping three days creates a pile-up of due cards; returning to a 100-card queue after a break feels daunting, and many learners never return. The problem is not just the backlog: cards that are reviewed late have already partially decayed, meaning the consolidation event is weaker than it would have been at the optimal moment.
The practical response to a backlog is not clearing it in one marathon session -- fatigue degrades encoding quality. Set a maximum session length of 15 to 20 minutes, work through the queue at that pace daily, and accept that some cards will be reviewed a day or two late. Returning to consistency matters far more than restoring a perfect schedule.
Rating recognition as retrieval
The subtlest failure mode is using spaced repetition in a way that trains recognition rather than retrieval. This happens when you glance at the answer before fully committing to a recall attempt, when card design includes too many context clues, or when you rate yourself as correct on a card you only partially retrieved.
The self-rating you give is the only input the algorithm has. Rating difficult cards as easy accelerates their intervals -- they disappear from the review queue precisely when they needed reinforcement. Be honest: if you saw the question and already knew what the answer would be without effortful recall, the interval may be too short and the card needs more review, not less.
What Memia does differently
Memia is not a promise of perfect memory. It is a tool that reduces the friction between wanting to learn something and actually doing the spaced repetition necessary to retain it long-term. The FSRS algorithm schedules each card at the moment that maximises retention per minute spent reviewing -- no manual sorting, no guessing when to revisit, no spreadsheet tracking.
Card creation is handled by AI: import a text, a PDF, describe what you want to learn, or browse the catalogue -- and Memia generates a structured deck with varied question formats (question-answer, multiple choice, true/false). Cards can be edited, split, or improved over time. A vague card becomes a clear one; a compound card becomes two focused ones.
The formats force retrieval rather than recognition: you see the question, the answer is hidden, and you generate your response before checking. Short daily sessions accumulate consolidation cycles. The goal is to build the kind of durable, accessible knowledge that transfers to real performance -- not knowledge that evaporates after the exam.
Memia is free to start. Create your first deck in under 5 minutes -- no prior flashcard experience required.
Frequently asked questions about spaced repetition
What is spaced repetition exactly?
Spaced repetition is a learning method that schedules reviews at increasing intervals based on how well you know each piece of information. Instead of reviewing everything at fixed times, the algorithm adapts: easy items come back after weeks or months; difficult ones come back after days. The key insight is that reviewing just before you forget -- not earlier, not later -- maximises the consolidation effect of each review.
What is the difference between spaced repetition and cramming?
Cramming compresses all reviews into a short period before an exam. It can produce short-term recall sufficient to pass the test, but that knowledge collapses quickly once the pressure ends. Spaced repetition distributes reviews over weeks and months, building progressively more stable memory traces. Research consistently shows spaced practice produces 2-3x better retention than massed practice at a 30-day horizon -- and the gap widens further at 6 months and beyond.
How much time per day does spaced repetition require?
5 to 15 minutes per day is enough for most learners maintaining an active deck of 200 to 500 cards. The key is consistency -- daily short sessions outperform occasional long ones because each session feeds an overnight consolidation cycle. If your sessions are taking 30 or more minutes regularly, you are probably adding new cards faster than your review pace can absorb. Reduce the daily new-card limit until sessions feel sustainable.
Does spaced repetition work for learning a language?
Yes -- vocabulary acquisition is one of the most studied and best-supported applications of spaced repetition, with strong evidence for 2-3x retention advantages over traditional study. The key is card design: test production (given an English word, produce the target-language equivalent) rather than only recognition. Spaced repetition builds the vocabulary foundation that makes all other language practice -- conversation, reading, listening -- more effective.
What is the difference between SM-2 and FSRS?
SM-2 (1987) uses a fixed ease-factor multiplier adjusted by performance scores. FSRS is a modern algorithm trained on millions of real review sessions that independently models memory stability and difficulty per card, producing intervals calibrated to each learner's actual forgetting rate. FSRS typically reduces over-review of easy cards and catches difficult cards at the right moment more reliably. Memia uses FSRS.
Can you do spaced repetition without an app?
Yes -- the Leitner box system (physical flashcards in numbered boxes, each reviewed at different frequencies) is the original pre-digital implementation and still works. However, apps handle scheduling automatically, track per-card performance, and adapt intervals based on your actual recall. The manual overhead of a physical system makes it significantly harder to maintain consistency long-term, especially with decks over 100 cards.
Is spaced repetition effective for exam preparation?
Very much so, for knowledge-heavy exams (medicine, law, languages, certifications). The key is starting early -- at least 4 to 6 weeks before the exam -- and adding cards incrementally as you cover new material. Spaced repetition is less suited for procedural skills or analytical reasoning, which require practice on actual problems. Use SRS for the factual foundation and problem sets for the reasoning layer that sits above it.