You have probably seen the claim everywhere: spaced repetition gives you 90% retention. Study blogs repeat it. Flashcard app landing pages repeat it. Reddit posts repeat it. But almost nobody explains where that specific number comes from — and most people who cite it are getting it subtly wrong.
The 90% figure is not a research finding extracted from Ebbinghaus or from meta-analyses on spaced practice. It is a default algorithm setting — a design choice made by the developers of FSRS, the modern spaced repetition algorithm used in Anki and other tools, to balance memory reliability against review workload. Understanding this distinction changes how you use spaced repetition entirely.
This article explains what spaced repetition retention 90 percent actually means, traces where the number came from, shows the real research behind it, works through the workload math for different targets, and gives you the concrete schedule that achieves it. If you are new to the method, start with our guide to spaced repetition study techniques first. If you are layering SRS on top of MOOC content or self-paced video lectures, the target retention math here pairs with our async study planning guide. If you already know the method and want to understand the 90% claim specifically, keep reading.
What 90% Retention Actually Means in Spaced Repetition
Before anything else, the definition needs to be precise, because the word “retention” is used loosely in most discussions and that causes real confusion.
In the context of spaced repetition algorithms like FSRS, retention (also called desired retrievability) refers to the probability that you can correctly recall a specific card at the moment it is due for review. A 90% retention target means the algorithm schedules each card so that, when it comes up for review, you have roughly a 90% chance of recalling it correctly.
This is not the same as:
- The percentage of cards you will remember forever
- Your overall deck accuracy at any random point in time
- The fraction of material you retain after completing a course
- Some guarantee that 90% of your memories are permanent
The distinction matters enormously. When people say “spaced repetition gives 90% retention,” they often mean the last item on that list — a claim about long-term memory permanence. That is not what the number measures. What it measures is scheduled recall probability at review time, which is a much more precise and actionable metric.
Think of it this way: if your retention target is 90% and you have 100 cards due today, you would expect to recall roughly 90 of them correctly on the first attempt. The 10 you miss get shorter intervals. The 90 you get right continue on their expanding schedule. The algorithm is continuously recalibrating so that each card arrives just before you are likely to forget it — specifically, at the point where your recall probability has dropped to your target threshold.
Understanding this also clarifies what “90% retention” in practice feels like: a normal session where you miss roughly 1 in 10 cards. Not an effortless perfect recall experience. The forgetting is intentional. It is the mechanism through which memory consolidates. Reviewing right at the edge of forgetting produces stronger long-term encoding than reviewing material you already remember well.
Where the 90% Number Came From (It’s Not Ebbinghaus)
Hermann Ebbinghaus published his forgetting curve research in 1885. His famous finding: without reinforcement, roughly 50% of new information is lost within 24 hours, and retention continues to decay exponentially from there. This research established that memory is fragile and time-dependent. What it did not produce is the 90% figure.
The specific 90% target comes from the FSRS algorithm (Free Spaced Repetition Scheduler), developed by Jarrett Ye (Expertium) and the open-source community around Anki. FSRS was designed to replace the older SM-2 algorithm with a more mathematically rigorous model of human memory based on two factors: stability (how long a memory persists) and difficulty (how hard a particular item is to recall). The default FSRS retention target shipped with the algorithm is 90%, and that default is what most learners inherit unless they explicitly change it.
When the FSRS team chose a default desired retention value, they ran analyses on what level maximized learning efficiency — the ratio of material retained per unit of review time spent. The analysis showed a clear sweet spot: above 90%, review frequency must increase sharply to maintain the higher threshold, producing rapidly escalating workload. Below 85%, intervals can be extended but recall becomes unreliable enough that you frequently re-encounter material as if it were new, defeating the purpose of the system. The 90% default sits at the bottom of the efficiency valley — high enough to be reliable, low enough to keep intervals manageable.
This is a calibrated engineering choice, not a measurement of how much you will remember in 10 years. The number could have been 85% or 92% — and Anki lets you change it to whatever you prefer. The reason 90% became the de facto standard is that it was the FSRS default, and FSRS became the most widely adopted modern spaced repetition algorithm. The convention spread from there.
The broader research on spaced repetition and retention — Cepeda et al., Kornell and Bjork, Roediger and Karpicke — demonstrates that distributed practice outperforms massed practice, that recall attempts strengthen memory, and that optimal gap lengths improve long-term retention. None of this research specifically validates 90% as a target. The research validates the method. The 90% target is a practical convention layered on top of that research.
The Research Behind Spaced Repetition Retention
While the 90% figure itself is a convention, the broader finding — that spaced repetition produces dramatically better retention than other study methods — is among the most replicated results in cognitive psychology. Here is what the actual studies show.
Cepeda et al. (2006, 2008) — The Definitive Spacing Effect Studies
Nicholas Cepeda and colleagues published two major papers that are widely cited as the empirical foundation for modern spaced repetition. In their 2006 meta-analysis of 254 studies covering 839 experimental conditions, they found that spaced practice consistently produced better long-term retention than massed practice (cramming) across a wide range of material types, subject populations, and retention intervals. This convergent evidence is the closest thing to a definitive spaced repetition retention 90 percent study — though, as noted, the 90% number itself is an algorithm choice rather than a direct experimental result.
Their 2008 paper went further, examining optimal gap lengths. A striking finding: at a 6-month retention test, subjects who studied with spaced intervals recalled approximately 95% of material, while subjects who used massed practice recalled approximately 37%. That is not a marginal effect. It is the difference between knowing something and not knowing it half a year later.
The study supporting 90% retention with spaced practice over massed practice is not a single paper but a convergent body of evidence. Cepeda 2008 is the most frequently cited because the comparison is stark and the retention interval is long enough to be practically meaningful.
Ebbinghaus (1885) — The Foundation, Not the Goal
Ebbinghaus’s original forgetting curve work established the baseline: without any reinforcement, memory decays rapidly. After 20 minutes, roughly 40% of new information is gone. After 24 hours, roughly 50–60% is gone. After a week, 70–80% is gone. After a month, more than 90% is gone for most types of material.
This work established the problem that spaced repetition solves. It is frequently mis-cited as evidence for the 90% retention target, but Ebbinghaus was documenting forgetting rates without intervention — the opposite of what spaced repetition achieves. Ebbinghaus also showed that re-learning forgotten material is faster than the initial learning (the “savings effect”), which prefigures why spaced repetition works: even when a memory fades below conscious recall, traces remain that accelerate re-learning.
Roediger and Karpicke (2006) — The Testing Effect
Henry Roediger and Jeff Karpicke demonstrated that retrieval practice (testing yourself) produces substantially better long-term retention than re-reading. In their experiment, students who repeatedly tested themselves retained 61% of material after one week, compared with 40% for repeated reading. The mechanism is the same as spaced repetition: the act of retrieval strengthens the memory trace in a way that passive review does not.
This research explains why flashcard review is more effective than highlighting and re-reading, and why the rating step in FSRS (Again/Hard/Good/Easy) matters: the difficulty of successful retrieval determines how much the memory is strengthened. Easy recalls produce less consolidation than effortful recalls. Reviewing material you already know perfectly is nearly wasteful. This is why the 90% threshold — which allows for roughly 10% difficulty — is more efficient than a 99% threshold.
For a deeper dive into what repetition actually changes at the memory level, see our article on what repetition actually trains in long-term memory.
The Math: 80% vs 90% vs 95% — Workload Curve
This is where the practical implications become concrete. Changing your retention target does not linearly change your workload — the relationship is exponential. A small increase in target retention produces a large increase in review frequency.
Here is the intuition: FSRS schedules the next review for the moment your retention probability would drop to your target. If your target is 90%, it waits until you have a 90% chance of remembering — then shows you the card. If your target is 95%, it shows you the card earlier, when you have a 95% chance — which means shorter intervals and more reviews. The higher the target, the less you forget between reviews, but the more often you must review.
In practical terms, the FSRS team’s analysis of their algorithm showed approximately the following relationship between retention targets and daily review counts for a mature deck of 1,000 cards:
| Retention Target | Approx. Daily Reviews | Relative Workload | Notes |
|---|---|---|---|
| 80% | ~90 reviews/day | 0.55× baseline | Lower effort, higher miss rate, frequent re-learning |
| 85% | ~115 reviews/day | 0.7× baseline | Reasonable for lower-priority material |
| 90% | ~165 reviews/day | 1× baseline (default) | FSRS default, efficiency sweet spot |
| 95% | ~330 reviews/day | 2× baseline | Roughly double the reviews vs 90% |
| 97% | ~530 reviews/day | 3.2× baseline | Likely unsustainable for most learners |
The numbers above are illustrative — actual review counts depend on deck composition, card difficulty, and your personal memory parameters. But the shape of the curve is real: raising target retention from 90% to 95% can roughly double the daily review burden. Raising it further to 97% more than triples it.
This is the core reason 90% became the standard: it is near the top of the range where you still get high reliability without the review load becoming prohibitive. An 80% target is manageable but produces enough misses that learning feels inefficient. A 95% target is achievable but requires a time commitment that most learners cannot sustain consistently.
There are legitimate cases for different targets. Medical students preparing for boards with high-stakes recall requirements might run a deck at 92–95% during an intensive study window. Language learners maintaining a large passive vocabulary might drop to 85% to keep reviews light. The key insight is that you are choosing a point on a trade-off curve, not selecting a fixed goal set by science.
The Real Protocol: A Schedule That Hits 90%
If you use an algorithm-based tool like FSRS (Anki, Flashcard Maker), the schedule is calculated automatically. But understanding the mechanics helps you interpret what the algorithm is doing and trust it enough to follow it consistently.
The FSRS Interval Ladder
For a new card rated “Good” consistently, FSRS produces a schedule roughly like this for a 90% retention target:
- After learning phase: 1 day
- Review 1: 3 days
- Review 2: 7 days
- Review 3: 21 days
- Review 4: 60 days
- Review 5: 180 days
- Review 6+: Continues expanding based on stability estimate
These are not fixed values — FSRS adapts each card individually based on your rating history. A card you rate “Hard” repeatedly gets shorter intervals. A card you rate “Easy” gets longer ones. The algorithm continuously refines its estimate of each card’s stability (how long it will persist in memory) and schedules accordingly.
What the Ratings Mean
- Again: You forgot or recalled incorrectly. The card resets to a short learning interval (minutes to hours). Stability estimate decreases.
- Hard: You recalled correctly but with significant difficulty. Interval grows by a small multiplier. Stability estimate increases modestly.
- Good: Normal successful recall. Interval grows by the standard multiplier. This is the expected rating for cards at your target retention level.
- Easy: You recalled with no effort. Interval grows by a larger multiplier. Use this when the card is clearly too easy — it pushes the next review further out.
The most common mistake is rating “Easy” too aggressively to reduce reviews, or rating “Good” when you genuinely struggled and should have rated “Hard”. Both distort the algorithm’s stability estimates and produce a schedule that no longer tracks your actual memory state. Rate honestly. The algorithm works better with accurate signal.
Is 90% Realistic With Daily Sessions?
Yes, with consistent daily sessions of 10–20 minutes. Here is the math for a realistic deck.
Assume you add 10 new cards per day. After 30 days, you have 300 cards in the system. At a 90% retention target with FSRS, a mature deck of 300 cards generates roughly 50–70 reviews per day — a session length of 10–15 minutes assuming around 10–15 seconds per card. After 90 days (900 cards), daily reviews grow to roughly 130–160, a 20–30 minute session.
The consistency requirement is real. The FSRS interval schedule assumes you review on schedule. If you miss three days, you return to a backlog. The algorithm handles backlogs gracefully — it reschedules overdue cards without inflating intervals inappropriately — but chronic skipping degrades effective retention below your target. The “90% is achievable” claim holds for learners who maintain daily or near-daily sessions. It does not hold for sporadic cramming.
For a full overview of the spaced practice protocol and how to build the daily habit, see our dedicated guide.
How to Track Your Actual Retention Rate
If you use Anki, your actual retention rate is visible in the statistics panel under Stats → Today or in the historical review log. The key metric is correct rate for mature cards (cards that have graduated from the learning phase). This is your actual retention rate in practice.
A few things to understand when reading these stats:
- Correct rate on young cards is lower than on mature cards. New material is inherently harder to recall. Mixing young and mature card stats produces a misleadingly low overall number.
- Correct rate on mature cards should track your retention target. If your FSRS target is 90% and your mature card accuracy is consistently 82%, the algorithm needs more of your review history to calibrate, or your actual memory parameters differ from the defaults.
- FSRS can be personalized. Anki supports FSRS parameter optimization based on your own review history. Running the optimizer produces parameters tuned to your specific memory characteristics, which will track your target more accurately than the defaults.
If you do not use Anki, track your correct rate manually by counting misses per session. 10 misses on 100 cards reviewed = 90% retention. Keep a simple log. Three weeks of data is enough to see whether you are above or below your target.
Common Mistakes That Tank Retention Below 90%
Even with a well-calibrated algorithm, several common behaviors systematically push real retention below the target.
Skipping Reviews
The single biggest driver of below-target retention is inconsistent review. The FSRS schedule assumes you review on the day a card is due. Every day you miss, your actual retention on those cards drops below your target. A backlog of 300 overdue cards does not represent a single missed session — it represents 300 individual memories that have decayed past their target retention level.
Cards That Are Too Large
Cards covering multiple facts are harder to rate accurately. If a card asks for three related facts and you remember two, do you rate it “Hard” or “Good”? This ambiguity produces noisy signal that confuses the algorithm. Atomic cards — one fact, one card — produce cleaner ratings and better scheduling. If your accuracy is chronically low on a specific card, the most likely cause is that the card is trying to test too much at once. Break it up.
Passive Reviewing
Looking at a card question and flipping to the answer without actually attempting recall is one of the most common and damaging habits. You are skipping the retrieval attempt — the mechanism that strengthens memory — and rating based on recognition rather than recall. Recognition is much easier than recall. Cards that feel easy to recognize may still fail under genuine recall conditions (a blank exam page, a real conversation). Always produce the answer mentally before flipping. Even an imperfect attempt is more valuable than no attempt.
Reviewing Buried Context
A card that was created from a specific textbook passage and contains implicit context cues (“in the chapter on homeostasis...”) will seem easier during review than it will be in an exam or real-world situation. Strip contextual cues from cards. The question should be answerable on its own, without reference to where it came from. This forces genuine recall rather than context-dependent recognition.
Overusing “Easy”
Clicking “Easy” feels efficient in the moment — it pushes cards far into the future. But if you mark something “Easy” when it was actually just familiar from a recent review, the algorithm schedules the next review too far out and you will miss it. Reserve “Easy” for cards you genuinely know cold, months after your last review. When in doubt, use “Good.”
More technique-level advice is covered in our guide to flashcard study techniques, which goes into card design, session structure, and common errors in detail.
Tools That Make 90% Easy (Without Hours of Drudgery)
The best tool for hitting 90% retention is one you will actually use consistently. That means low friction for both card creation and daily review.
Anki + FSRS
Anki is the reference implementation for FSRS and the most configurable spaced repetition tool available. You can set a custom desired retention percentage, run parameter optimization on your review history, and inspect statistics at a granular level. If you want full control and are willing to spend time on setup, Anki is the answer. See our Quizlet alternatives comparison for a broader look at where Anki fits relative to other tools.
Flashcard Maker (Chrome Extension)
Flashcard Maker is a free Chrome extension with FSRS built in. It solves the card creation friction problem: highlight any text on any webpage, right-click, and the card is created and scheduled — without leaving the page. All reviews happen in the Chrome side panel. Storage is local via IndexedDB, no account required, works offline. Import Quizlet TSV or CSV decks. Export your decks to a Quizlet-ready TSV file.
For learners who do most of their reading in a browser — documentation, research papers, online textbooks — this eliminates the context switch that causes most people to never create the card in the first place. The default retention target is 90%, and the FSRS schedule runs automatically. You rate Again/Hard/Good/Easy after each card and the algorithm handles the rest.
See our overview of the best flashcard apps for a full comparison of tools across different study workflows.
Choosing Between Tools
If you study from physical books, scanned PDFs, or non-browser sources, Anki is the better fit. If most of your learning happens online — reading articles, documentation, or web-based textbooks — Flashcard Maker removes the largest barrier to consistent card creation. Both tools run FSRS and can achieve the same 90% retention target. The difference is workflow, not algorithm.
For language learners specifically, both tools work well. Our guide to SRS for language learning covers how to structure vocabulary decks, set realistic targets, and integrate spaced repetition into an immersion workflow.
Frequently Asked Questions
Why is 90% retention the standard for spaced repetition?
It is not a research-derived universal standard — it is the FSRS algorithm’s default desired retrievability value, chosen because it sits at the efficiency sweet spot between high recall reliability and manageable review workload. At 90%, the review burden is roughly half of what it would be at 95%, while still keeping recall probability high enough to feel reliable. The convention spread because FSRS became the dominant modern spaced repetition algorithm.
What is the ideal spaced repetition schedule?
With FSRS at 90% retention, a well-rated card follows an expanding interval schedule: roughly 1 day, then 3 days, then 7 days, then 21 days, then 60 days, then 180 days and beyond. The exact values depend on card difficulty and your rating history. FSRS adjusts each card individually — there is no single universal schedule. The pattern of expanding intervals is consistent; the specific gaps are personalized.
How much do you actually retain with spaced repetition?
At a 90% target, you will correctly recall approximately 90% of cards on the day they are scheduled for review. Long-term retention of well-established cards (those that have reached multi-month intervals) is high — but it depends on continued review. Spaced repetition is a maintenance system as much as a learning system. Stop reviewing a deck entirely and the memories will decay. The Cepeda et al. 2008 study showed ~95% recall at 6 months with spaced practice; maintaining that requires continuing the schedule past 6 months.
Can you achieve higher than 90% with spaced repetition?
Yes. You can set FSRS to target 92%, 95%, or higher. The cost is approximately doubled review workload at 95% compared to 90%, and roughly tripled at 97%. Whether that trade-off is worth it depends on how critical recall is for your specific use case. For most learners, 90% is the right default. For high-stakes material (medical licensing, bar exam, aviation certification, the digital SAT — see our SAT vocab list — or AP exams (our AP CSP study timeline walks through this), a short-term increase to 93–95% during an intensive study window can be justified.
What is the difference between spaced repetition retention and learning retention?
Spaced repetition retention (desired retrievability) is the probability of correct recall at a scheduled review time. “Learning retention” is an informal term often used to mean how much of a course or subject you remember after completing it — a much fuzzier concept. Spaced repetition improves both, but the 90% number only specifically describes the former.
Does the 90% retention claim come from Ebbinghaus?
No. Ebbinghaus documented forgetting rates without intervention, which showed roughly 50% retention after 24 hours and steeper decay over subsequent days and weeks. His work established the problem that spaced repetition addresses. The 90% figure comes from the FSRS algorithm design, not from Ebbinghaus or from any single research study.