Spaced Repetition: A Complete, Practical Guide

What is spaced repetition?

That is the whole idea. Everything else — algorithms, apps, ease factors, retention targets — is machinery for answering one question well: when should I see this next?

The practical shape of it looks like this. You have a collection of questions, usually flashcards. Each day the system hands you the ones that are due. You try to answer each from memory, check, and tell the system how it went. Cards you found easy get pushed further into the future. Cards you fumbled come back soon. Over weeks, a well-run deck settles into a state where most cards sit at intervals of months and your daily queue stays small even as the deck grows.

The counterintuitive part is that the difficulty is the point. A review that feels effortless taught you almost nothing. A review where you paused, strained, and then produced the answer did real work. This is why spaced repetition consistently beats rereading despite feeling worse while you do it — a mismatch we come back to below, because it is the single biggest reason people abandon the method.

Why spacing works

Storage strength and retrieval strength

Robert and Elizabeth Bjork's new theory of disuse distinguishes two things people usually lump together. Storage strength is how thoroughly a memory is embedded — it is assumed only to grow. Retrieval strength is how accessible that memory is right now, and it decays.

Immediately after you read something, retrieval strength is very high and storage strength is low. That is the cramming state: you can produce the answer instantly, and it will be gone in a week. The theory's key claim is that the increase in storage strength from a successful retrieval is larger when retrieval strength has fallen. In plain terms: the harder you had to work to find the answer, the more the act of finding it helped.

Spacing is simply a way of arranging for retrieval strength to have dropped a useful amount before you test yourself again — but not so far that you fail outright and learn nothing.

Two other contributors

• Encoding variability. Study the same thing on Tuesday morning and the following Saturday evening and you encode it in two different contexts — different mood, different room, different surrounding thoughts. More retrieval routes lead to the memory. Massed study gives you one route.
• Consolidation. Memories are stabilised over hours and days, with sleep playing a substantial role. Spacing reviews across days means each repetition acts on a memory that has already been partly consolidated, rather than on a still-fragile trace.

What the research actually shows

Three results are worth knowing in a little detail, because they determine how you should actually set things up.

Cepeda and colleagues, 2006. A meta-analysis of the distributed practice literature — 254 studies, 317 experiments, over 14,000 participants. The spacing advantage held broadly. It is the single best answer to "is this a real effect or a study-hack fad".

Cepeda and colleagues, 2008. The one that matters for scheduling. Over 1,300 participants learned obscure trivia facts, then reviewed after a gap ranging from minutes to 105 days, then were tested after a delay ranging from 7 to 350 days. Two findings:

• The optimal gap between study and review grew with how far away the final test was. There is no universally correct interval — the right one depends on when you need the material.
• As a proportion of the retention interval, the optimal gap shrank as the delay grew. Very roughly, for a test a week away the best first gap is around a day; for a test a year away it is on the order of weeks, not months.

The advantage of spacing also widened as the final test moved further into the future. Spacing does not just help; it helps most exactly where you care most.

Dunlosky and colleagues, 2013. A review in Psychological Science in the Public Interest assessed ten common study techniques for utility. Practice testing and distributed practice were the only two rated high. Rereading, highlighting and summarisation — the three things most students actually do — were rated low.

Two caveats an honest guide has to include. First, much of this research uses relatively simple materials — word pairs, trivia facts, short passages — over weeks rather than the years a medical student cares about. The direction of the effect is not in doubt; the exact size for your material is less certain. Second, "low utility" does not mean useless. Rereading is a reasonable way to encounter material the first time. It is a poor way to commit it to memory, which is a different job.

The condition nobody mentions

This is the most common way people run spaced repetition badly, and it is invisible from the outside — your streak looks identical either way.

Karpicke and Roediger's 2008 study in Science makes the point sharply. Participants learned Swahili–English word pairs under four conditions, varying whether learned items were dropped from further study or from further testing. A week later, items that continued to be tested were recalled around 80% of the time. Items dropped from testing but still restudied fell to roughly a third. Repeated studying, once the item had been learned once, added essentially nothing. Repeated testing did almost all the work.

So the practical rule is: before you flip a card, commit to an answer. Say it aloud, or at least form it fully in your head. If nothing comes after eight or ten seconds, that is a legitimate failure — mark it as such and move on. Sitting and staring at a card for two minutes is neither retrieval nor study; it is just discomfort.

The mechanics of doing this properly are the subject of a separate guide: active recall — what it is and how to actually do it.

How to choose your intervals

The 10–20% heuristic

Cepeda's 2008 data is often compressed into a usable rule of thumb: the first review should come after roughly a tenth to a fifth of your retention interval. It is a simplification — the ideal proportion falls as the delay stretches — but it is far better than a fixed ladder.

Why fixed ladders break down

Consider two cards you made on the same day. One asks for the capital of Peru. The other asks you to distinguish two Japanese particles that behave almost identically in half their uses. A fixed schedule gives them the same treatment. In reality the first should be at a six-month interval within a few weeks, and the second will need a dozen reviews before it survives a fortnight.

An algorithm that watches your answers handles this automatically. It pushes the easy card out fast, keeps the hard one close, and — critically — keeps your total daily workload roughly constant while doing so. This is the entire value proposition of a scheduler, and it is why "which algorithm" is a question worth caring about: see FSRS vs SM-2.

The retention dial

Modern schedulers let you set a target — the probability you want of recalling a card when it comes due. The usual default is 90%. It is worth understanding the trade-off, because it is not linear:

• Raising the target to 95% means shorter intervals and considerably more reviews for a modest gain in accuracy.
• Lowering it to 85% cuts your daily workload noticeably, at the cost of failing a few more cards — and each failure is itself a useful, if effortful, learning event.

If your queue is unmanageable, lowering the retention target is a more sensible response than skipping days. If you are preparing for a specific high-stakes exam in the near term, raising it for the final month is reasonable.

How many cards a day

Here is why the multiplier exists. Every new card generates a stream of future reviews: a few in the first week, a couple in the first month, then progressively fewer as its interval stretches into months. Add a new card every day and those streams overlap into a steady load. The load stabilises — this is the good news — but it stabilises at a level set entirely by your new-card rate.

Treat these as order-of-magnitude figures. The multiplier depends on your retention target, how well your cards are written, and how hard the material is — well-formed cards on familiar material land at the bottom of each range, and badly-formed cards on unfamiliar material can blow past the top of it.

The classic failure is to set 50 new cards a day in week one, feel great for nine days, and then meet a 400-card queue that you never recover from. The daily cost of a new card arrives weeks after you add it. Start at 10 and raise it only after a month at steady state, when you can actually see the bill.

When it goes wrong

Three failures account for nearly every abandoned deck.

The backlog

You travel for a fortnight. You return to 900 due cards and quietly stop opening the app.

Understand what a backlog actually is: not damage, just a bill. Overdue cards have decayed further than intended, so you will fail more of them than usual, and the scheduler will respond by shortening their intervals. That is the system working. Nothing is broken.

The recovery that works:

• Cap your daily reviews at something you will genuinely do — say 20% above your normal load — and clear the backlog over a week or two rather than in one heroic session.
• Do not add new cards until the backlog is gone. This is non-negotiable; adding new cards to a backlog is pouring water into a leaking bucket.
• Consider lowering your retention target temporarily. It lengthens intervals across the board and reduces how fast the queue refills.
• If the backlog is enormous — say, more than a month of accumulation and thousands of cards — seriously consider deleting the cards you no longer need rather than reviewing them out of guilt. A deck you actually use is worth more than a complete one you don't.

Cards that never stick

Some cards fail again and again. Anki calls these leeches and flags them after eight lapses; the concept applies whatever you use. A small number of cards will consume a wildly disproportionate share of your review time.

The instinct is to try harder. That almost never works, because the problem is nearly always the card, not you. The usual culprits:

• It asks for too much. "Describe the stages of mitosis" is an essay, not a card. Split it.
• It interferes with a sibling. Two cards for two similar things, each of which reminds you of the other. You are not failing to remember; you are failing to keep them apart. Rewrite them so each front contains the feature that distinguishes them.
• You never understood it. Memorising something you cannot follow is the most expensive way to study. Go and understand it, then rebuild the card.

Delete or rewrite; don't grind. How to make flashcards that actually work covers the rewrites in detail.

It feels like it isn't working

This is a feature, not a bug, and it is worth naming because it is the reason many people quit while the method is succeeding.

Roediger and Karpicke's 2006 experiments found that students who studied by testing themselves substantially outperformed students who reread — but that the rereading group predicted they would do better. Fluency feels like knowledge. Rereading is fluent. Retrieval is not. Your sense of how well a session went is close to worthless as a measure of learning, and is often inverted.

The fix is to stop trusting the feeling and look at the data instead. A deck with a stable or rising success rate on mature cards is working, regardless of how effortful yesterday's session felt.

What spaced repetition is bad at

The honest limitations, since almost nobody lists them:

• It cannot generate understanding. Comprehension has to come first — from a lecture, a textbook, a worked example, a conversation. Spaced repetition preserves what you understood; it does not manufacture it.
• It is weak for procedures. Integration by parts, differential diagnosis, writing a proof — these need practice under varied conditions. Cards can hold the components (which substitution, which cut-off value) but not the skill of choosing.
• Transfer is limited. A memory practised as a specific question tends to be strongest in response to that question. Vary your card formats and include "why" and "when" cards alongside "what" cards, or you build knowledge that only answers to one phrasing.
• Garbage in, garbage forever. The method is indifferent to whether a card is worth knowing. It will faithfully drill you on a badly-worded card for years. Curation matters more than volume.
• The opportunity cost is real. Forty-five minutes of daily reviews is forty-five minutes not spent doing practice problems or reading. For subjects with a large factual base — languages, medicine, law — the trade is clearly worth it. For a maths course, cards should be a small supplement to problem sets, not a replacement.

A 30-minute setup

Concretely, if you are starting today:

1. Pick one subject. Not three. The habit is the hard part, and it forms around one deck.
2. Set your new-card limit to 10 per day. You will want more. Don't.
3. Make 20 cards from material you already understand. One fact per card, a question on the front that has exactly one correct answer. Starting with familiar material means your first week is easy, which is how habits survive.
4. Choose a fixed time and attach it to something you already do — coffee, the commute, before bed. "When X, then reviews."
5. Review every day for two weeks before changing anything. Resist the urge to tune settings you don't yet have data about.
6. At two weeks, look at your stats. If your success rate on review cards is above roughly 85–90% and the session takes less time than you allotted, raise your new-card limit. If not, leave it and improve your cards instead.

Frequently asked questions

Where this comes from