Forgetting Curve

Definition:

The forgetting curve is a mathematical model of how memory strength decays over time after initial learning. First described by Hermann Ebbinghaus in 1885, it shows that memory loss is rapid immediately after learning and slows progressively — following a pattern well described by an exponential or power function. Each successful spaced repetition review resets and stabilizes the curve, which is why SRS algorithms are built around it.

Also known as: Ebbinghaus curve, memory decay curve, retention curve, curve of forgetting

In-Depth Explanation

Ebbinghaus quantified forgetting by learning lists of nonsense syllables (to strip away the effect of prior associations), waiting varying lengths of time, and then measuring how much less effort it took to re-learn the list compared to learning it from scratch — the “savings” method. What he found was that a fixed proportion of memory is lost per unit of time early on, but the rate of loss slows as time passes. Memory does not decay linearly; it decays fastest immediately after learning and then stabilizes.

The approximate retention $R$ at time $t$ after learning is often expressed as:

$$R(t) = e^{-t/S}$$

where $S$ is the stability of a given memory — how resistant it is to forgetting. A high-stability memory (a word you’ve reviewed 10 times over months) decays very slowly; a low-stability memory (a word you first saw today) decays fast. This two-variable model — current retrievability and memory stability — is exactly what FSRS models for each card, and why FSRS outperforms SM-2‘s simpler approach.

The critical insight for SRS design is that each successful review does two things simultaneously:

It resets retrievability to near 100% (you just recalled it — it’s fresh again).
It increases stability — the memory now decays slower than it did before.

This compounding effect is why SRS intervals can grow from 1 day ? 3 days ? 10 days ? 1 month ? 6 months for well-learned items. The forgetting curve flattens with each review. Conversely, a failed recall does not just reset the interval — it signals that stability was overestimated, and the algorithm recalibrates.

Individual forgetting curves vary substantially. Ebbinghaus studied himself, under artificial conditions (nonsense syllables), with no interfering material. Real-world vocabulary learning involves prior associations, emotional salience, contextual cues, and interference from similar items — all of which shift the curve. This is why modern algorithms like FSRS derive per-card stability parameters from individual user data rather than using fixed values.

Common Misconceptions

“Forgetting is bad and should be prevented.”

Counterintuitively, some forgetting is necessary for optimal retention. Reviewing an item before any forgetting has occurred produces a weaker consolidation effect than reviewing it when it is on the edge of being forgotten. This is the mechanism behind retrieval practice — the slight difficulty of recall at the forgetting threshold is what strengthens the memory. SRS doesn’t try to prevent forgetting; it tries to catch items just as they’re about to be forgotten.

“Everyone has the same forgetting curve.”

Ebbinghaus studied his own memory under controlled conditions. Individual forgetting rates vary significantly based on age, sleep quality, emotional state, material difficulty, prior knowledge, and attention during initial learning. This is why adaptive algorithms (FSRS) that learn from individual user data outperform fixed-interval systems.

“Once you forget something completely, it’s gone.”

Research on relearning consistently shows that items that appear to be completely forgotten are re-acquired substantially faster than items learned for the first time — evidence that partial memory traces persist below conscious recall. This is the “savings” Ebbinghaus himself measured.

“The forgetting curve is the same for meaningful and meaningless material.”

Ebbinghaus deliberately used nonsense syllables to create a “pure” forgetting curve. Meaningful material — vocabulary in context, narrative information, emotionally salient content — decays slower because it has more associative hooks that support retrieval. Real-world SRS forgetting is therefore gentler than the Ebbinghaus curve implies.

History

1885: Hermann Ebbinghaus publishes Über das Gedächtnis, presenting the first empirical forgetting curve derived from his own self-experiments with nonsense syllables. His methodology — measuring savings in relearning time as a proxy for residual memory — becomes the template for memory research. [Ebbinghaus, 1885]

Early 1900s: Replications of Ebbinghaus’s curve confirm exponential decay across different material types, learner populations, and delay intervals, establishing it as one of the most robust findings in psychology.

1932: C.A. Mace draws practical learning prescriptions from the forgetting curve, recommending distributed review timing in The Psychology of Study. [Mace, 1932]

1972: Sebastian Leitner‘s Leitner System uses the forgetting curve concept practically — cards are reviewed at increasing intervals, with harder items returning to shorter cycles, mimicking what the curve predicts. [Leitner, 1972]

1985–1990: Piotr Wozniak builds SuperMemo explicitly on the forgetting curve, using his own data to derive interval-stability relationships that became SM-2. [Wozniak, 1990]

2015: Murre and Dros publish the first controlled modern replication of Ebbinghaus’s forgetting curve experiments, confirming the original shape and decay rate with contemporary neuropsychological methodology. [Murre & Dros, 2015]

2022: FSRS uses a formal two-component memory model (retrievability R and stability S) to model the forgetting curve per-card per-learner, rather than using the fixed formulas of SM-2. [Ye, 2022]

Criticisms

Ebbinghaus’s original forgetting curve research has been criticized for its extremely limited experimental scope — all data came from Ebbinghaus himself memorizing nonsense syllables, producing ecological validity concerns when the results are applied to meaningful material, naturalistic learning, or the diversity of learner populations. The specific mathematical parameters of the curve (including the decay constants used in spaced repetition algorithms) were derived from these minimal artificial conditions and may not accurately reflect forgetting rates for meaningful L2 vocabulary, complex grammatical structures, or real-world communicative competence. More recent research has proposed that forgetting follows different trajectories for different types of knowledge and different levels of initial encoding strength.

Social Media Sentiment

The forgetting curve is one of the most widely referenced findings in language learning communities and productivity/learning social media — it directly motivates spaced repetition systems (SRS) and the “review before you forget” advice that is foundational to tool recommendations for vocabulary learning. The forgetting curve illustration routinely appears in explainer content about why languages are forgotten without practice and why SRS works. While few community members engage with the detailed mathematics or Ebbinghaus-specific concerns, the basic insight — that retention decays predictably and that timely review resets the curve — is widely understood and applied.

Last updated: 2026-04

Practical Application

Understanding the forgetting curve suggests two core study habits: space reviews before the predicted forgetting threshold (not after complete forgetting), and ensure robust initial encoding — shallowly learned material has a steeper forgetting curve than deeply processed material.

Related Terms

Research

Ebbinghaus, H. (1885/1913). Memory: A Contribution to Experimental Psychology. Teachers College, Columbia University.
Summary: The original source of the forgetting curve. Ebbinghaus’s nonsense syllable experiments established that memory decays exponentially and that relearning effort quantifies residual memory. The foundational reference for all SRS scheduling theory.

Murre, J.M.J., & Dros, J. (2015). Replication and analysis of Ebbinghaus’ forgetting curve. PLOS ONE, 10(7), e0120644. https://doi.org/10.1371/journal.pone.0120644
Summary: A modern controlled replication of Ebbinghaus’s original experiments confirming the shape, rate, and exponential nature of the forgetting curve using contemporary methodology. Establishes that the original findings hold under rigorous modern conditions.

Wozniak, P.A. (1990). Optimization of learning [Master’s thesis]. University of Technology, Poznan. https://www.supermemo.com/en/archives1990-2015/english/ol
Summary: Wozniak translates the forgetting curve into algorithmic SRS scheduling — deriving the interval formulas for SM-2 from empirical data on his own learning. The direct applied bridge between Ebbinghaus’s theoretical curve and modern SRS practice.

Bahrick, H.P., Bahrick, L.E., Bahrick, A.S., & Bahrick, P.E. (1993). Maintenance of foreign language vocabulary and the spacing effect. Psychological Science, 4(5), 316–321.
Summary: Demonstrates the forgetting curve in real-world foreign language vocabulary and shows that spaced review dramatically reduces decay rate. Directly relevant to SRS design for language learners — confirms that the Ebbinghaus curve applies to meaningful vocabulary, not just nonsense syllables.

Smolen, P., Zhang, Y., & Byrne, J.H. (2016). The right time to learn: Mechanisms and optimization of spaced learning. Nature Reviews Neuroscience, 17(2), 77–88. https://doi.org/10.1038/nrn.2015.18
Summary: Provides a neurobiological explanation for the forgetting curve — protein synthesis windows, synaptic tagging, and consolidation timing. Explains why the time between reviews matters at a cellular level and why SRS-timed reviews align with biological memory consolidation processes.