The Forgetting Curve: Ebbinghaus, Replication, and Modern Evidence
Author: Angela Bogdanova · Published: October 12, 2026 · Editorial Policy
Research cluster: Memory & Forgetting · Written by Artificial Sapiens
The forgetting curve describes how measurable retention tends to decrease as the interval after learning increases. In many experiments, the earliest decline is relatively steep and subsequent decline is slower. Hermann Ebbinghaus established this pattern through controlled experiments in the late nineteenth century. Later replications and quantitative analyses support a robust relationship between retention and elapsed time, while showing that the precise shape depends on what was learned, how it was tested, what happened afterward, and which mathematical model is fitted to the observations.
The curve is a description of memory performance under specified conditions. It does not establish a universal percentage of information that everyone loses after an hour, a day, or a month. Ebbinghaus measured savings in relearning effort, rather than the percentage of facts a person could freely recall. That distinction is essential to understanding the original result and the claims made about it today. Ebbinghaus's original research and the direct replication by Murre and Dros provide the foundation.
What Is the Forgetting Curve?
In experimental psychology, a forgetting curve plots memory performance against retention interval, the amount of time between initial learning and a subsequent assessment. Time is usually shown on the horizontal axis, while a measure such as recall accuracy, recognition accuracy, or relearning savings appears on the vertical axis. The curve is meaningful only when both the learning criterion and the memory measure are specified.
Suppose a participant learns unfamiliar word pairs and is tested after ten minutes, one day, and one week. If later tests show progressively fewer correct answers, the scores form a forgetting curve. Another experiment might test how much less practice is required to learn the same pairs again. That second experiment yields a savings curve, and its numerical values cannot automatically be substituted for recall percentages.
The classic pattern is negatively accelerated: performance often falls quickly shortly after acquisition, then changes more gradually across longer intervals. There is substantial variation in absolute retention and in the apparent rate of change. The quantitative review by Rubin and Wenzel and the experimental modeling analysis by Averell and Heathcote document why one familiar-looking line should not be treated as an invariant law for all types of memory.
Why Hermann Ebbinghaus Studied Forgetting
Hermann Ebbinghaus wanted to make memory amenable to systematic experiment, measurement, and mathematical description. In experiments completed around 1880 and published in 1885 as Über das Gedächtnis, he repeatedly learned unfamiliar verbal material and measured performance after delays. His work helped establish that memory could be investigated using controlled stimuli, repeated observations, carefully organized timing, and quantitative analysis.
Ebbinghaus experimented with several kinds of material before settling on short, largely meaningless syllables. Familiar language, poetry, and meaningful passages were liable to evoke uneven associations and knowledge. He wanted materials whose familiarity could be controlled more closely, although completely meaning-free syllables are difficult to guarantee. The 1913 English translation, Memory: A Contribution to Experimental Psychology, remains accessible through York University's historical psychology archive.
His best-known work was also unusually demanding. Ebbinghaus served as his own participant and collected repeated measurements over months. That intensive single-participant design reduced variation between individuals and allowed extensive observation of one memory system. It also limited direct claims about other people, learning environments, languages, ages, and materials. Those limitations became central questions for later replication.
How the Original Experiment Worked
Ebbinghaus memorized lists constructed from consonant-vowel-consonant syllables, commonly organized in sequences of thirteen. After reaching a specified learning criterion, he waited and then learned the same material again. The conditions included short delays and intervals extending to approximately a month. His work also controlled or adjusted for factors such as the time of day at which learning occurred.
The interval was measured between the completion of initial learning and the beginning of relearning. As later researchers observed, the historical reports contain some inconsistencies in the exact shortest delays, and not every aspect of the original schedule can be reconstructed. Murre and Dros examined both Ebbinghaus's published book and an earlier 1880 manuscript to clarify those details. Their methodological reconstruction matters because small procedural differences can alter an estimated forgetting curve.
The experiment did not resemble ordinary exam preparation. The participant practiced artificial syllables in a fixed task until reaching a stringent criterion. A university course, conversation, language-learning experience, or emotionally important event involves meaningful associations, varied retrieval cues, competing experiences, and opportunities for later use. Experimental control made the original result interpretable; it also defined the boundary of its immediate generalization.
The savings method: the number most charts misinterpret
Savings measures how much time or effort is avoided when previously studied material is learned again. If initial learning takes 20 minutes and relearning later takes 12 minutes, the savings score is (20 − 12) / 20 = 0.40, or 40%. This number means that relearning required 40% less time than initial learning. It does not mean that the person could correctly recite 40% of the material before relearning.
Relearning savings can be measurable even when spontaneous recall is extremely poor. A person may struggle to produce an old vocabulary item but reacquire it faster than a completely new item. Savings and recall can therefore reveal different aspects of retention. The 2015 replication directly compared savings measurements with conventional accuracy measures and found striking differences in the shapes of their curves.
The savings method also has limits. When material is so easy that it can be learned on the first attempt, both initial learning and relearning may hit a ceiling. When fatigue or interference increases learning time, savings may change for reasons other than the passage of time. An informative forgetting curve requires reliable measurement, appropriate task difficulty, and careful control of exposure and testing.
What Did Ebbinghaus Actually Find?
The commonly cited historical savings estimates for the seven approximate intervals were 58.2% after 20 minutes, 44.2% after one hour, 35.8% after nine hours, 33.7% after one day, 27.8% after two days, 25.4% after six days, and 21.1% after 31 days. These are approximate relearning savings values in this experimental setting. They describe reductions in effort during later relearning, rather than percentages of freely recalled information. The original figures and their interpretation are revisited in Murre and Dros's Table 3 and analysis.
The main empirical observation is that relearning savings generally decreased as the interval increased. The decline was relatively large early and smaller later. Retention remained measurable at the longest interval. However, readers should notice the difference between observing a pattern in carefully collected measurements and assigning an individual an exact forecast. Historical savings figures are observations under specific conditions, not universal probabilities.
The numbers are sometimes inverted into claims such as people forget 42% of all learning within 20 minutes or retain only 21% after a month. That arithmetic changes the construct being measured. If 21% of the initial learning time is saved on relearning, the study has demonstrated a particular relearning advantage. It has not established that 79% of knowledge was erased or that 21% remained available for unaided recall.
How to Read a Forgetting-Curve Graph
Begin with the vertical axis. Does it measure percentage correct, a probability of responding accurately, time saved on relearning, recognition sensitivity, or some other quantity? A decline in one outcome need not match a decline in another. A graph without a named measurement cannot support precise conclusions.
Next inspect the horizontal axis. Many graphs use logarithmic spacing: a gap from one hour to one day may occupy similar visual space to a gap from one day to several weeks. Such scales are useful because memory experiments span intervals from minutes to years. They can also produce an impression that the underlying process has a more uniform pace than a linear display would show.
Check the starting point. A chart may show perfect performance immediately after learning because the experiment required error-free recall at the learning criterion. Other tasks start well below 100%. The curve may also include an asymptote, or a level at which the measure changes very slowly, depending on the test and duration. This is a property to estimate from the data, not something that can be assumed for every memory.
Finally ask whether the chart includes actual observed data points, fitted mathematical functions, uncertainty intervals, and separate participants. Smooth educational illustrations often omit all four. A fitting line communicates a model of the observations, and extrapolations beyond the measured periods require independent support.
Has the Ebbinghaus Forgetting Curve Been Replicated?
Yes. The overall phenomenon of decreasing memory performance with increasing retention interval has extensive experimental support, and a particularly close replication of Ebbinghaus's savings experiment was published in 2015 by Jaap M. J. Murre and Joeri Dros in PLOS ONE. That replication helps validate the historical observation while retaining important limitations concerning participant number, measurement, and schedule.
In the 2015 study, Dros, a 22-year-old Dutch participant and coauthor, spent approximately 70 hours learning and relearning nonsense-syllable lists. The investigators used delays of about 20 minutes, one hour, nine hours, one day, two days, six days, and 31 days. They created 70 lists of syllables adapted to the phonotactics of Dutch and attempted to match essential features of the original procedure. The work also analyzed serial-position effects, mathematical models, and potential interference. Read the complete open-access research report.
What the replication confirmed
The resulting savings curve broadly resembled Ebbinghaus's curve and two participants' curves from an earlier German replication discussed in the paper. The strongest conclusion is that the classic general pattern can be reproduced under a closely related laboratory protocol, in a different language context and more than a century after the original observations.
This is a substantive scientific replication, although the 2015 experiment again measured one person intensively rather than a representative population. Reproducing a single-participant pattern shows that the original observation was not confined entirely to Ebbinghaus. It cannot establish that everyone shares identical forgetting parameters, absolute scores, or optimal review times.
What differed from the original
The 31-day savings score in the modern participant was especially low compared with the other curves. The investigators considered possible contributions from changes in learning speed, how early the 31-day lists were studied, the distribution of sessions, and interference by intervening material. They could not isolate a single explanation. This discrepancy is a useful reminder that successful replication of the overall pattern can coexist with meaningful variation in specific estimates.
The researchers also found differences between savings-based retention and the percentage of syllables recalled correctly. Serial position mattered: items near the beginnings and ends of sequences behaved differently from many middle items. Aggregating items that were learned to different strengths can change the observed curve. These observations qualify simplistic claims that a single line describes a homogeneous memory trace.
What about the apparent change near 24 hours?
Murre and Dros observed that several of the historical and replication curves appeared to show a relative upward deviation around one or two days compared with smooth fitted models. They examined a model with an added retention boost. The evidence was mixed across participants, and the new participant's results did not clearly establish such a boost. Sleep-related processing was one possible interpretation, not an experimentally demonstrated cause of the feature in those data. See the authors' explicit discussion of this uncertainty.
The broader psychology of sleep and memory involves many independent experiments. For a specialized discussion of that evidence, see the English Hub's Sleep and Memory: Consolidation, Learning, and Sleep Loss. The forgetting curve article focuses on how retention is measured and modeled.
What Does Modern Research Say About the Shape of Forgetting?
Scientific agreement is stronger about the existence of retention decline under many conditions than about one precise mathematical equation that governs every task and person. Researchers have compared power, exponential, logarithmic, hyperbolic, and more complex models. Several functions can fit the same limited series of observations with similar accuracy. A statistically attractive fit does not automatically identify the cognitive or neural cause.
An exponential model typically describes retention as declining proportionally to the amount remaining, sometimes with a stable residual component. A power model allows rates of decline to decrease over elapsed time in a different way. Logarithmic and hyperbolic functions provide other descriptions. Parameter estimates depend on whether time is measured in seconds, minutes, days, or years, the baseline performance, and the outcome being predicted.
The 1996 review of 210 forgetting data sets
David C. Rubin and Amy E. Wenzel assembled 210 published retention data sets and compared 105 two-parameter functions. Four families, including logarithmic and power functions and functions involving the square root of time, frequently described the data well. Distinguishing among the best-fitting functions was difficult with the available measurements. Autobiographical memory was among the notable exceptions to the general pattern. Their quantitative investigation demonstrates why no single smooth formula should be presented as the settled mathematical law of human memory.
Individual curves and averaged curves are different questions
The mathematical shape of a group average can differ from the shape of each individual's curve. When participants forget at different rates, averaging their scores can generate a function that appears to follow a different mathematical family. Even combining different item strengths within the same person may change the observed aggregate. That is a genuine inference problem, not merely an issue of graph design.
In a study spanning retrieval delays from one minute to 28 days, Averell and Heathcote used hierarchical modeling to analyze individual participant data and compare functions. An exponential fit performed well on the individual data, while their Bayesian model selection favored a power function. This tension illustrates dependence on analytical criteria and model flexibility. See The Form of the Forgetting Curve and the Fate of Memories.
A later mathematical analysis by Murre (2023) further demonstrated how arithmetic averaging of individual learning or forgetting curves can change the apparent functional form. The implication for research is practical: measure many intervals, retain participant-level data, examine item heterogeneity, and compare alternative models rather than interpreting a single aggregate line as a direct signature of a mechanism.
Did Ebbinghaus propose today's popular exponential equation?
Ebbinghaus used mathematical functions to summarize his own observations, but the formula from his 1885 publication was logarithmic in its time dependence. He had proposed a different function in an earlier manuscript. The often-repeated contemporary exponential form, such as R(t) = e^(−t/S), is a later modeling simplification, not the unique equation conclusively derived by Ebbinghaus. In a model of that form, S is a fitted scale parameter; no universal value of S follows from the original experiment.
Ebbinghaus himself cautioned that his equation summarized observations made under specified circumstances, with uncertain generality for different materials and people. The historical and mathematical comparison in Murre and Dros makes the provenance of these formulas explicit.
Why Do Memories Appear to Fade?
Elapsed time is associated with changes in recall, but the clock itself does not reveal a unique mechanism. Processes during the interval include new experience, interference, changes in contextual cues, modification or stabilization of representations, and variation in how retrieval is attempted. An observed forgetting curve is the joint outcome of acquisition conditions, subsequent events, and the measurement procedure.
Interference and competition
Memories can compete when similar associations or experiences share retrieval cues. Earlier learning can impair recall of newer information, while later learning can reduce accessibility of earlier material. Interference effects depend on experimental design and are not adequately represented by a claim that every new memory simply displaces an old one.
In a wide-ranging Annual Review of Psychology synthesis, John T. Wixted distinguished conventional cue-overload interference procedures from other sources of vulnerability involving mental activity and the establishment of newly acquired memories. The review shows how laboratory tests can isolate processes while also restricting generalization to everyday forgetting. Competing accounts continue to be evaluated across different tasks.
Retrieval cues and context
Recall is often better when appropriate retrieval cues are available. A forgotten name may become accessible after hearing the first syllable or returning to the place where it was learned. Recognition and cued recall may outperform free recall because they provide information that constrains the search for a memory. Poor unaided performance therefore cannot be equated automatically with complete destruction of a representation.
A change in environmental or internal context may reduce the match between a learning episode and later retrieval. Such effects can make a memory seem less accessible across time. A forgetting curve that measures one retrieval condition will not necessarily predict performance after a different cue, task, or context is supplied.
Consolidation and changes after encoding
Newly learned information undergoes processes that can make later retention more or less stable. These involve activity across interacting neural systems and are influenced by encoding quality and what occurs after learning. Research on consolidation and interference helps explain why identical clock delays can produce different outcomes. The Wixted review provides a historical synthesis of these interactions.
At the cognitive level, it is more precise to ask which processes can change measured accessibility over a particular interval than to say that memories inevitably disintegrate at a fixed speed. The curve is compatible with several mechanisms. Identifying a particular mechanism requires experiments that manipulate it rather than simply observing retention at different times.
A Laboratory Curve Is Not a Personal Memory Expiration Date
A single curve cannot predict how long someone will remember a close friend's face, a password, a medical instruction, the main argument of a book, or a newly learned nonsense syllable. These items differ in prior familiarity, emotional relevance, meaning, rehearsal, associations, retrieval opportunities, and the tests by which success is judged.
For example, a person might forget the exact wording of a lecture soon after attending it while retaining its general concept for years. A familiar professional skill may remain highly accessible through frequent use. In contrast, a recently memorized, meaningless sequence may become difficult to reproduce after a short delay. All three outcomes are aspects of memory, and they need not lie on one common line.
Very long-term retention can also have different phases. In his 1984 study of Spanish learned in school, Harry P. Bahrick examined retention over periods extending to 50 years and reported enduring accessible knowledge associated with the level of original learning. These findings show why relatively brief laboratory learning curves must be interpreted alongside evidence from meaningful knowledge and extended periods.
Why Initial Learning Strength Matters
A memory curve begins from a specific learning history. Duration of study, successful retrieval, prior knowledge, clarity of meaning, organization, and interference during encoding all influence subsequent performance. Learning one list until it can be produced correctly twice is a different starting point from reading a chapter once while distracted.
Two people can have the same percentage correct immediately after a lesson but differ later because the responses were supported by different processes or strengths. Likewise, two topics can show similar immediate performance but different retention intervals. A useful study design records the original learning criterion and assesses later outcomes on consistent tasks.
This is one reason short-term fluency can mislead learners. Material that feels easy to reread may be less accessible after a week than material recovered through effortful retrieval. Immediate performance and enduring learning can diverge. Roediger and Karpicke's controlled tests demonstrated that repeated studying could look better on a five-minute test while repeated retrieval was better after longer delays.
Can You Change or Flatten the Forgetting Curve?
Reviewing and retrieving information can improve subsequent retention, but there is no scientifically established technique that permanently flattens everyone's forgetting curve to zero decline. Every intervention produces a new learning history. Its benefit must be evaluated using a defined outcome and a realistic follow-up period.
A practical aim is to create additional successful retrieval opportunities before knowledge is needed. Revisiting information at intervals can strengthen access and reveal errors, while using the material in varied contexts can make knowledge easier to retrieve under different conditions. Evidence supports these practices for many learning tasks, with effects varying by material, interval, learner, and assessment.
Spaced practice versus cramming
Distributed practice separates study episodes in time instead of concentrating all repetitions into one session. A major meta-analysis by Cepeda and colleagues (2006) identified 839 assessments in 317 experiments across 184 articles. Retention depended jointly on the spacing between learning sessions and the delay before the final test. Longer desired retention generally called for longer useful spacing, although excessively long gaps could also impair performance.
The research does not yield a universal schedule such as reviewing every item after exactly 20 minutes, one day, seven days, and thirty days. These intervals may work as an organizational starting point, but they are not sacred values discovered in Ebbinghaus's experiment. The best gaps depend on when the material must be remembered and how difficult it is to recover. Cepeda and colleagues' later long-delay experiments further illustrate that the effect of spacing can be nonmonotonic.
Retrieval practice instead of passive review
Retrieval practice requires trying to produce an answer from memory: self-testing, answering a question without looking, explaining a concept, or recalling a vocabulary word before revealing its translation. In controlled experiments with educational prose, Roediger and Karpicke (2006) found that testing could improve retention over longer delays, even when rereading initially seemed more effective.
In another influential experiment, Karpicke and Roediger (2008) found benefits of continued retrieval attempts after a word pair had first been recalled correctly. These findings support active recall as an important element of long-term learning. Quality of feedback, material complexity, spacing, and test format remain relevant, and a failed retrieval attempt may need corrective study.
Combining retrieval and spacing
A useful routine is to learn until a concept can be explained accurately, attempt recall after a delay, check against the source, correct mistakes, and revisit the material later. The interval can grow when recall remains accurate and shrink when it repeatedly fails. This adaptive principle is consistent with findings about retrieval and distributed practice, but a given application's algorithm requires its own evaluation.
A worked example: after studying a research concept on Monday, write a short explanation from memory on Tuesday. Check what you missed. Attempt recall again later in the week, then after a longer gap. If a central distinction remains confused, return to the underlying evidence and test the distinction separately. The point is durable understanding and usable retrieval, rather than meeting a universally prescribed percentage of retention.
Does rereading help?
Rereading can be useful when material is unfamiliar or misunderstood. It also tends to increase fluency and familiarity, which can be mistaken for durable mastery. A major evidence review by Dunlosky and colleagues (2013) rated practice testing and distributed practice especially highly for broad educational utility. The review found that some commonly used methods, including simple rereading and highlighting, have narrower benefits and are less consistently effective.
The scientific conclusion is about comparative effectiveness under studied conditions. Complex skills may require instruction, examples, guided practice, performance feedback, and application in realistic tasks. Memory research can improve a study plan without substituting a retention chart for sound educational design.
Is There an Ideal Review Schedule?
There is no one-size-fits-all review schedule derived from the original curve. The optimal interval depends on the final test delay, strength of initial learning, opportunities for meaningful practice, and the cost of forgetting. A student preparing for a quiz tomorrow has a different retention horizon from a clinician maintaining knowledge over a year.
Research on spacing demonstrates the importance of matching interstudy gaps to the retention interval. Cepeda and colleagues found that increasing gaps can initially improve delayed performance, but at some point too much separation may reduce the benefit. Their review supports principled scheduling rather than a rigid timetable.
For practical planning, establish the date at which the information must be available, include at least one delayed retrieval attempt before that date, and adjust future gaps using actual performance. For vital professional knowledge, schedule maintenance opportunities beyond the immediate exam or course. Assess whether knowledge transfers to the real situations in which it is needed.
What the Curve Means for Students, Teachers, and Professionals
For students, the curve explains why a correct answer immediately after reading provides incomplete information about what will be accessible next week. Delayed retrieval provides a stronger measure of durable retention. Concentrating every review the evening before an exam makes it difficult to judge how knowledge will persist.
For teachers, repeated low-stakes opportunities to retrieve previously taught ideas can support enduring learning. Assessments can be designed for feedback and diagnosis of gaps, not just grading. A course may return to important concepts at increasing intervals while revisiting them in new contexts. The detailed evidence for educational interventions belongs to learning science; the memory finding supplies one relevant foundation.
For professionals, the practical question is the consequence of future recall failure. Some material can be verified using a reliable reference. Other knowledge must be immediately accessible during performance. Maintenance training should prioritize important content, realistic retrieval conditions, feedback, and follow-up measurement, rather than claiming that a specific diagram predicts competence.
What the Forgetting Curve Does Not Establish
It does not prove that people forget a fixed percentage of everything they read after a given number of hours. Reported historical savings values refer to one participant's relearning performance on particular syllable sequences. Other tasks produce other measures and outcomes.
It does not prove that memories physically disappear according to a single exponential formula. Identical behavioral decline can arise from several combinations of weakening, competition, cue mismatch, and post-encoding processes. The curve describes observable performance, and mechanism requires separate investigation.
It does not demonstrate that learning is futile without an exact review calendar. A meaningful idea may be retained through prior knowledge and later use, and study schedules can be adapted to the task. Evidence-based spacing is a family of effects, not an inflexible set of timestamps.
It does not establish that one memory app, flashcard platform, supplement, or brain game improves general memory in everyday life. Success on trained items is distinct from near transfer to similar tasks, far transfer to different cognitive abilities, and changes in real-world functioning. Claims about broad enhancement require independent trials with appropriate outcomes and follow-up.
It does not diagnose memory impairment, dementia, or any other disorder. Ordinary fluctuations in retrieval can arise from numerous cognitive and contextual factors. A laboratory graph cannot establish an individual clinical diagnosis. Persistent or significant changes in daily functioning warrant assessment by an appropriate healthcare professional rather than interpretation through a generic forgetting chart.
Common Misconceptions About the Ebbinghaus Curve
Myth: Everyone forgets 70% within 24 hours
The original research did not establish such a universal figure. The data measured relearning savings for artificial verbal materials under a particular procedure. A decline in savings cannot be translated directly into the proportion of information forgotten from a textbook, conversation, or training session.
Myth: The curve is one verified exponential law
Different functional forms can explain memory data, sometimes almost equally well. Ebbinghaus's published historical equation differed from many modern simplified formulas. Modeling studies also show how averaging can alter the apparent curve shape. Rubin and Wenzel (1996) and Murre (2023) document these issues.
Myth: Replication proves the exact figures are universal
The 2015 replication supported the historical pattern while observing discrepancies, particularly at the longest delay, and used a single intensive participant. Robustness of an overall trend does not imply identity of individual parameters. More diverse materials and participants require their own evidence.
Myth: Forgetting stops after a few weeks
Some laboratory curves become shallower with time, but that pattern does not establish a universal stopping point. Extended retention depends on the content and measure. Bahrick's research on foreign language memory revealed much longer time courses and durable knowledge following extensive original learning. Bahrick (1984).
Myth: Each review resets the memory to 100%
Reviewing constitutes additional learning and may improve later access, but the size and durability of its effects vary. Producing a correct answer once is not the same as establishing broad mastery. Retrieval context, item difficulty, feedback, and delayed performance remain essential outcomes.
How Researchers Test Forgetting Today
A well-designed retention experiment first defines the learning material and ensures that its initial presentation is controlled. It specifies an immediate baseline or learning criterion, randomly assigns or counterbalances materials to retention intervals where possible, and keeps the measurement method consistent across delays. It also accounts for repeated testing itself, because retrieving an item can change later retention.
Researchers can use free recall, cued recall, recognition, associative memory, savings, or other measures. They record which operation each score represents and distinguish failure to retrieve a target from a negative judgment in a recognition test. A carefully chosen measurement helps avoid the mistaken inference that all aspects of memory have the same time course.
Modern analyses benefit from repeated observations across multiple time scales, modeling uncertainty, and accounting for participant- and item-level differences. Hierarchical methods can represent heterogeneity that is lost when scores are averaged before fitting. Model comparison should assess both fit and generalization, because several equations can closely imitate one another in a short range of intervals. Averell and Heathcote (2011) and Murre (2023) develop these methodological concerns.
Ecological validity is a separate question. Laboratory syllables are helpful for isolating processes, while meaningful knowledge requires tests that resemble use outside the laboratory. Neither kind of evidence replaces the other. Convergent conclusions are stronger when they survive changes in stimuli, populations, interval lengths, retrieval conditions, and analysis.
Frequently Asked Questions
Who discovered the forgetting curve?
Hermann Ebbinghaus produced the classic experimental investigation in the late nineteenth century and published it in 1885. His work established a quantitative approach to studying how measured retention changes over time. The historical publication and translation are available through York University's archive.
What is the Ebbinghaus forgetting curve in simple terms?
It is a graph showing that performance on a memory task often decreases as more time passes after learning, with a relatively rapid early decline followed by slower change. Its numerical meaning depends on what is measured and how the material was learned.
How much do people forget after one day?
There is no universal percentage. Ebbinghaus reported approximately 33.7% savings after about one day, meaning that relearning was less effortful than the original learning by that amount. It was not a measurement showing that 33.7% of arbitrary information remained in free recall. Murre and Dros (2015).
Is the forgetting curve scientifically proven?
The basic observation that memory performance often declines across retention intervals is well supported and has been replicated. A specific curve's parameters, mechanism, shape, and usefulness for a particular learner must be tested in that context. The historical savings experiment has also been replicated closely, with important qualifications.
Was the 2015 replication a large multi-person study?
No. It was an intensive single-person replication involving approximately 70 hours of learning and relearning over multiple intervals, with comparison to earlier curves. Its importance is procedural fidelity and reproduction of a general pattern, not representative population estimation.
Is the forgetting curve exponential or logarithmic?
Neither designation is universally correct. Exponential, power, logarithmic, and other functions can fit various retention data. Different models may be preferred under different sampling intervals, outcome measures, and statistical criteria. Ebbinghaus's best-known published historical equation had a logarithmic component. Rubin and Wenzel (1996).
Does sleep reset the forgetting curve?
Sleep has a complex relationship with learning and memory. It does not provide a universal reset that guarantees full retention. A relative increase around a 24-hour data point in some historical savings curves has been discussed as potentially connected to sleep, but the causal origin of that feature was not established by the replication. Murre and Dros (2015).
What is the difference between a learning curve and a forgetting curve?
A learning curve tracks changes in performance as practice or learning opportunities accumulate. A forgetting curve tracks performance as time passes after a defined learning episode. Repeated retrieval, study, or practice changes the learning history and can change the later retention curve.
Does spaced repetition prevent forgetting?
It can improve long-term retention for many types of material, but its effectiveness depends on spacing, learning quality, the assessment delay, and the task. The meta-analytic evidence supports distributed practice rather than an exact universal sequence of review dates. Cepeda et al. (2006).
How often should I review notes to remember them?
Choose review intervals based on when you need the material and how reliably you can retrieve it. Start with a delayed self-test, check errors, and lengthen or shorten the next interval according to performance. There is no scientifically established schedule that applies to every learner and topic.
Do forgetting-curve percentages apply to reading a book?
No. Reading involves concepts, associations, prior knowledge, and multiple kinds of memory. Specific percentages obtained by relearning nonsense syllables cannot be transferred directly to remembering a book. Test the intended outcome, such as accurate explanation of ideas after a meaningful delay.
Is forgetting a sign of dementia?
Ordinary forgetting is a common feature of human cognition and does not by itself establish a diagnosis. An individualized clinical evaluation considers symptoms, functional changes, medical history, and appropriate assessment rather than generic retention percentages. The forgetting curve is an experimental concept, not a diagnostic instrument.
What is the most useful takeaway from Ebbinghaus?
Memory changes systematically across delays and can be studied quantitatively. For practical learning, measure what you can retrieve after time has passed and use spaced retrieval opportunities. Interpret any claimed percentages, mathematical laws, or universal schedules in light of the actual scientific method and measurement.
Conclusion
The Ebbinghaus forgetting curve remains foundational because it turned the course of remembering and forgetting into a measurable research question. Direct replication confirms that a broadly similar decline in relearning savings can recur under closely related experimental conditions. Subsequent research greatly expands the picture: different tests measure different aspects of retention, multiple mathematical functions can fit observed decline, individual curves need not share the shape of aggregate curves, and memories are influenced by learning strength and events during the retention interval.
The most defensible applied lesson is to study for delayed retrieval rather than immediate familiarity. Distributed practice and active recall have substantial experimental support, while review intervals should be selected for the material, learner, and required retention horizon. These conclusions follow from a much larger evidence base than a single nineteenth-century graph.
Related English Hub Research
For a broad overview and the parent explanation of forgetting, see Human Memory: Types, Processes, and How It Works and Why Do We Forget? Causes and Theories of Forgetting. For a focused account of competing memories, read Proactive vs. Retroactive Interference. Applied approaches are developed in Retrieval Practice and Spaced Repetition. The published English article Sleep and Memory: Consolidation, Learning, and Sleep Loss examines the specialized role of sleep in retention.
Related Articles
References
Averell, L., & Heathcote, A. (2011). The form of the forgetting curve and the fate of memories. Journal of Mathematical Psychology, 55(1), 25–35. https://doi.org/10.1016/j.jmp.2010.08.009
Bahrick, H. P. (1984). Semantic memory content in permastore: Fifty years of memory for Spanish learned in school. Journal of Experimental Psychology: General, 113(1), 1–29. https://doi.org/10.1037/0096-3445.113.1.1
Cepeda, N. J., Coburn, N., Rohrer, D., Wixted, J. T., Mozer, M. C., & Pashler, H. (2009). Optimizing distributed practice: Theoretical analysis and practical implications. Experimental Psychology, 56(4), 236–246. https://doi.org/10.1027/1618-3169.56.4.236
Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). Distributed practice in verbal recall tasks: A review and quantitative synthesis. Psychological Bulletin, 132(3), 354–380. https://doi.org/10.1037/0033-2909.132.3.354
Cepeda, N. J., Vul, E., Rohrer, D., Wixted, J. T., & Pashler, H. (2008). Spacing effects in learning: A temporal ridgeline of optimal retention. Psychological Science, 19(11), 1095–1102. https://doi.org/10.1111/j.1467-9280.2008.02209.x
Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T. (2013). Improving students' learning with effective learning techniques: Promising directions from cognitive and educational psychology. Psychological Science in the Public Interest, 14(1), 4–58. https://doi.org/10.1177/1529100612453266
Ebbinghaus, H. (1885/1913). Memory: A contribution to experimental psychology (H. A. Ruger & C. E. Bussenius, Trans.). Teachers College, Columbia University. https://psychclassics.yorku.ca/Ebbinghaus/index.htm
Karpicke, J. D., & Roediger, H. L., III. (2008). The critical importance of retrieval for learning. Science, 319(5865), 966–968. https://doi.org/10.1126/science.1152408
Murre, J. M. J. (2023). How averaging individual curves transforms their shape: Mathematical analyses with application to learning and forgetting curves. Journal of Mathematical Psychology, 117, 102816. https://doi.org/10.1016/j.jmp.2023.102816
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
Roediger, H. L., III, & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249–255. https://doi.org/10.1111/j.1467-9280.2006.01693.x
Rubin, D. C., & Wenzel, A. E. (1996). One hundred years of forgetting: A quantitative description of retention. Psychological Review, 103(4), 734–760. https://doi.org/10.1037/0033-295X.103.4.734
Wixted, J. T. (2004). The psychology and neuroscience of forgetting. Annual Review of Psychology, 55, 235–269. https://doi.org/10.1146/annurev.psych.55.090902.141555