memori
The short answer
Ten new cards a day settles at about 100 reviews a day. That is 15–20 minutes once your deck reaches steady state, and it is a good default for almost everyone. If you want a single number to start with and no further thinking: set your new-card limit to 10, leave your review limit uncapped, and revisit in a month.
The reason the question is usually asked the wrong way round is that "how many should I review" is not a setting. Reviews are a consequence. The setting is how many new cards you introduce, and everything else follows from it with a delay long enough that most people never connect cause to effect.
The multiplier nobody tells you
At steady state, your daily review count is roughly 10× your daily new-card count. The multiplier comes from how many times an average card is reviewed before its intervals grow long enough to stop mattering — typically eight to twelve times over the first couple of years, concentrated in the first few months.
Where the number comes from is worth seeing, because it lets you sanity-check any claim about workload.
Follow one card through a typical schedule: it comes back after 2 days, 8 days, 3 weeks, 7 weeks, 4 months, 10 months, 2 years. That is 7 reviews in the first three and a half years, but 4 of them land in the first three months. Add lapses — most decks fail somewhere between 10% and 20% of reviews, each of which adds two or three extra reviews to that card — and the average card costs you somewhere around 8–12 reviews before it goes quiet.
So:
| New cards/day | Steady-state reviews/day | Minutes/day at 8s per card | Deck after 1 year |
|---|---|---|---|
| 5 | ~50 | 7 | 1,800 |
| 10 | ~100 | 15 | 3,650 |
| 15 | ~150 | 20 | 5,500 |
| 20 | ~200 | 27 | 7,300 |
| 30 | ~300 | 40 | 11,000 |
| 50 | ~500 | 65+ | 18,000 |
Two things people find surprising in that table. The first is how fast the deck column grows — 20 new cards a day is over seven thousand cards a year, which is more than most people have any real use for. The second is how sane the minutes look even at the top, which is misleading: 65 minutes on a good day is 65 minutes on a bad day too, and the bad days are what determine whether a habit survives.
The multiplier moves with your retention target, your lapse rate, and your card quality. A deck with clean atomic cards and a 0.85 target might run at 7×. A deck full of compound cards at a 0.95 target can hit 15× or worse, because failed cards re-enter the queue repeatedly. If your observed ratio is far above 10×, the diagnosis is almost always card quality rather than settings.
What the first six weeks look like
The delay between cause and effect is what makes this hard to learn by experience. Here is the shape at 15 new cards a day, which is a common and reasonable starting choice:
| Day | New | Reviews due | Total | Time |
|---|---|---|---|---|
| 1 | 15 | 0 | 15 | 3 min |
| 3 | 15 | ~20 | 35 | 6 min |
| 7 | 15 | ~45 | 60 | 9 min |
| 14 | 15 | ~85 | 100 | 14 min |
| 21 | 15 | ~120 | 135 | 19 min |
| 35 | 15 | ~140 | 155 | 21 min |
| 60+ | 15 | ~150 | 165 | 22 min |
Week one is delightful. Week two is fine. Somewhere around day 16 to day 22 the load roughly doubles from where it was a week earlier, and this is where people conclude the app is broken or that they have made a terrible mistake. Neither is true — the reviews you scheduled in week one are simply arriving, on top of the ones from week two, on top of the new cards you are still adding.
After about six weeks it flattens. The flattening is real and it is the whole promise of the method: the deck keeps growing but the daily cost stops growing, because mature cards leave the rotation faster than new ones enter it.
Knowing the spike is coming is most of the defence against it. If you want the full week-by-week version with what to do at each stage, that's your first 30 days with spaced repetition.
Start from minutes, not cards
Decide how many minutes a day you will genuinely spend on a bad day — tired, busy, travelling — and derive the card count from that. Most people can answer 6–10 cards a minute on mature material. Twenty sustainable minutes is roughly 150 reviews, which implies a new-card limit of about 15.
The reason to work in minutes is that minutes are what you actually have and cards are not. But there is a second reason: card counts are not comparable across subjects, and minutes are.
| Card type | Realistic time each | Cards in 20 minutes |
|---|---|---|
| Vocabulary recognition (word → meaning) | 3–5 s | 250–400 |
| Vocabulary production (meaning → word) | 6–8 s | 150–200 |
| Definitions, dates, single facts | 5–8 s | 150–240 |
| Cloze deletions in context | 8–12 s | 100–150 |
| Multi-step reasoning, derivations | 30–90 s | 15–40 |
That bottom row is the one worth pausing on. If your cards require working something out rather than retrieving it, the 10× multiplier still applies but your minutes budget buys an order of magnitude fewer cards. A hundred derivation cards a day is a two-hour session. People studying maths and physics who set limits by copying a language learner's numbers discover this painfully.
The corollary: if your session is far longer than the card count suggests, your cards are too big. That is a card-writing problem with a card-writing fix — split them.
Starting numbers by situation
| Situation | New/day | Settles at | Note |
|---|---|---|---|
| Curious, no deadline | 5–10 | 50–100 reviews | Under 15 min. Easy to keep for years. |
| Language learner, serious | 15–20 | 150–200 | Cards are fast, so this stays around 25 min. |
| Undergraduate, one heavy course | 10–15 | 100–150 | Add cards weekly from lectures, not in bulk before exams. |
| Professional exam, 6–12 months out | 20–30 | 200–300 | Sustainable only if you've already got the habit. |
| Medical school, dedicated study | 40–60 | 400–600 | An hour-plus daily, and a genuine commitment. |
| Returning after a gap | 0 | — | Add nothing until the backlog is cleared. See below. |
The last row is the one most often ignored and it is the most important. Adding new cards while carrying a backlog is like taking on new debt to service old debt — the queue never converges and every day feels like failure.
Those figures are real — large shared decks and 40–60 new cards a day are normal in that context — but they come with conditions that rarely get stated alongside them. The cards are usually pre-written and heavily refined by thousands of users, the material is unusually well suited to retrieval practice, and the students have restructured their week around it. Copying the numbers without the conditions is how people end up with a 500-card queue and a deck they wrote themselves in a hurry.
Working backwards from a deadline
If you have a fixed exam date, the calculation runs the other way, and there is a constraint people miss.
Say you have 900 cards to learn and 100 days. Naïvely that's 9 new cards a day. But cards introduced in the last two weeks will have had one or two reviews at most, which is not enough for them to be reliable on the day. In practice you want every card introduced at least three weeks before the exam, which means you have 80 usable days, not 100 — so 12 a day, settling at about 120 reviews.
The general form:
- New cards per day = total cards ÷ (days until exam − 21)
- Check the result against the 10× rule. 12/day → ~120 reviews → ~18 minutes. Sustainable.
- If the implied review load exceeds your available minutes, you have too many cards, not too little time. Cut the deck. Something has to give, and it is better for it to be low-value cards than the whole habit.
- Stop introducing new cards entirely in the final 7–10 days and just review. The last week is for consolidation, not acquisition.
If the deadline is close enough that this arithmetic produces an absurd number — 300 cards in 10 days — then spaced repetition is the wrong tool for that deadline, and you should read spaced repetition vs cramming instead, which covers what to actually do with one day left.
Fixing a limit you set too high
The standard situation: you set 50 new cards a day in week one because it felt slow, and now you have 600 due and a growing sense of dread. This is fixable and does not require deleting anything.
- Set new cards to zero. Today. Not "a bit lower" — zero, until the queue is genuinely under control. This is the only step that actually stops the bleeding.
- Lower your desired retention to about 0.85 if your scheduler supports it. This lengthens every interval in the deck at once and can take 20–30% off the daily count without you reviewing anything.
- Cap daily reviews at a number you'll actually finish — say 100 — and let the overdue cards drain over two or three weeks. A capped queue you clear every day is psychologically completely different from an uncapped one you never do.
- Delete without guilt. Go through the backlog and remove cards you no longer need, cards you can't answer because they're badly written, and duplicates. A deck assembled in enthusiasm usually has 10–20% that is pure liability.
- Restart new cards at a third of your old limit once the backlog is clear, and only after a full week of clearing it.
Recovery takes two to four weeks for most backlogs. The failure mode is doing steps 1 and 3 for three days, feeling better, and turning new cards back on — which puts you back where you started by the end of the month.
The other dial: desired retention
If your scheduler uses FSRS, desired retention is the second lever on workload and it acts on your entire deck at once. Moving from 0.90 to 0.85 lengthens all intervals and can cut daily reviews by 20–30%, at the cost of a few more lapses. Moving to 0.95 can nearly double the load for a small accuracy gain.
The workload curve is steeply asymmetric around the default. This is why "review less" is usually better achieved by lowering the target than by skipping days: skipping days does not reduce the work, it defers it and adds lapses on top.
A reasonable policy for most people: sit at 0.90 by default, drop to 0.85 during a heavy term or when the queue is unsustainable, and raise to 0.95 only for the final three or four weeks before something high-stakes — then put it back. The mechanics are covered in FSRS vs SM-2.
Whatever app you use, find its statistics screen and look at two things: your true retention rate, and your review count over the last month. Memori shows both, along with the daily breakdown of new versus review cards — which is the view that makes the ramp above obvious rather than mysterious. Most of the decisions on this page are guesswork without that data and straightforward with it.
Signals you have it right
Rather than chasing a number, check these four things after a month at whatever limit you chose:
- You finish the queue on most days, including the bad ones. Not every day — but if you're skipping more than one day a week, the limit is too high.
- Your true retention sits between 80% and 92%. Below 80% and the material is either too hard, badly carded, or being introduced faster than it can consolidate. Above 92% and you are reviewing things you already know — you can afford to add more, or lower the target and free the time.
- The daily count has stopped rising after six to eight weeks at a constant new-card limit. If it's still climbing at week ten, your lapse rate is high enough to be worth investigating.
- You don't dread it. This sounds soft and it is the most predictive of the four. The dominant failure mode in spaced repetition is not inefficiency, it is abandonment, and volume is the usual cause.
A smaller deck reviewed daily for two years beats a larger deck reviewed for six weeks and abandoned, by an enormous margin. Every setting on this page should be chosen with that in mind.
Frequently asked questions
How many flashcards should I review per day?
You don't choose the review count directly — you choose a new-card limit, and reviews follow about three weeks later at roughly ten times that rate. Ten new cards a day settles at around 100 reviews, or 15–20 minutes. For most people starting out, 10–15 new cards a day is the right range.
How many new cards a day is too many?
Above about 30 a day is unsustainable without a professional reason, since it implies roughly 300 daily reviews within a month. Medical students on 40–60 are accepting an hour or more daily and have usually restructured their week around it. If you're not in that situation, treat 20 as a ceiling.
Why did my reviews suddenly double in week three?
Because cards from week one are returning for their third and fourth reviews while you're still adding new ones, and the arrival rates stack. It's the most predictable event in spaced repetition and the most common point at which people quit. Nothing is broken — the load is catching up to the limit you set on day one.
How long should a daily session take?
A length you'd still complete on a bad day — for most people 15–25 minutes. At 6–10 seconds per card that's roughly 100–200 reviews. A session you'd skip when tired is too long, because consistency contributes more to the outcome than volume does.
Should I review all my due cards every day?
Yes, if the number is sustainable. If it isn't, lower your new-card limit and retention target so fewer cards come due, rather than leaving due cards unanswered. Partially clearing the queue each day builds a backlog that compounds; lowering the inputs fixes it permanently.
Does the number depend on the subject?
Less than people expect. What varies is time per card and interference between similar items. Vocabulary is quick and supports high volume; multi-step reasoning cards are slow, so the same minutes buy far fewer. Set the limit from available minutes rather than from someone else's card count.
Is it better to do one long session or several short ones?
Several short ones, if the choice is available — it's easier to sustain, and breaking the queue into two or three sittings spreads the retrievals slightly, which is marginally better than doing them back to back. The effect is small; the adherence benefit is not.
What if I have a huge deck I imported rather than built?
The same arithmetic applies, but the deck size is now fixed and only the new-card limit controls how fast you meet it. A 10,000-card shared deck at 20 new a day takes 500 days to introduce fully. Decide up front whether you want all of it — imported decks are usually 30–40% material you have no use for.
Where this comes from
- The Anki manual, Deck Options — new card limits and review limits, for how limits interact and why capping reviews behaves differently from capping new cards.
- Open Spaced Repetition. FSRS specification and desired-retention workload analysis. The source of the claim that workload rises steeply above a 0.90 target.
- Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). Distributed practice in verbal recall tasks. Psychological Bulletin, 132(3), 354–380. For the underlying spacing effect that makes the interval growth in the multiplier calculation possible.
- 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. On distributed practice and practice testing as the two highest-utility techniques.
- Review-count and interval figures in the tables are modelled from typical FSRS behaviour at a 0.90 retention target with a 10–15% lapse rate. They are illustrative rather than measured, and your own statistics screen is the authority for your deck.
