The Forgetting Curve: Does Failed Recall Mean Nothing Remains?

You cannot recite a sequence you once knew. Has learning lost every effect? An important move in Hermann Ebbinghaus's memory research was not to ask only whether he remembered. He examined how much work it took to learn the same material again. If relearning required less effort than the original learning, something from the earlier experience could still matter without being freely recited.

The forgetting curve grew from that measure of relearning. It is often presented as a timetable specifying identical percentage losses for everyone. Reading the original investigation reveals a different account: a particular person studying particular material under defined conditions, then discussing results and sources of error. Understanding the curve requires looking at the operations behind the descending line. [1–2]

Why reduce meaning?

Ebbinghaus constructed syllables with a vowel sound between consonants and combined them into lists. Roughly 2,300 syllables supplied material for selection and rearrangement. He wanted to reduce variations in meaning, emotion and association found in prose or poetry. A striking narrative passage and an ordinary descriptive sentence were difficult to treat as equivalent learning units. [1, §§11–12]

The material does not represent all everyday memory. It narrows the process under study, allowing features such as list length and repetition to be varied more clearly. Artificial conditions do not make an experiment worthless; they identify the question it tries to answer with control. Transferring its result to everyday texts requires remembering that narrowing.

Nonsense was not perfectly uniform

Ebbinghaus acknowledged that the syllables were not ideally homogeneous. Some lists were easier than others. Seeing, hearing and speech movements participated together. Familiarity with particular sound combinations in a native language could also produce differences. [1, §12]

“Nonsense” therefore does not mean entirely neutral material that creates no other effects. Reducing meaning is not removing every variation. Reading a study requires considering which differences have been reduced and which remain. The things being learned contribute uncertainty to the measurement itself.

Defining when learning ends

In the forgetting experiments, learning continued until two errorless recitations were possible. After an interval, the same material was relearned to that criterion. The initial and later endpoints were thus intended to be comparable. Instead of simply saying that studying occurred, the experiment specified what counted as completion. [2, §27]

That matters because comparing material initially learned to very different levels would answer another question. Yet a criterion cannot solve everything. Ebbinghaus noted that the first successful recitation sometimes disappeared immediately afterward and that brief changes in attention could alter performance. A measurable endpoint does not guarantee an identical mental state in each trial. [1, §14]

A second question after time has passed

The investigation used intervals including approximately twenty minutes, one hour, nine hours, one day, two days, six days and 31 days. Time required for original learning was compared with time required to learn again. The main question concerned returning to a learning criterion, rather than simply counting syllables that came to mind. [2, §27]

This savings measure separates failed visible recall from the disappearance of every learning effect. If less work is needed later, the previous encounter contributes. That contribution is not a photograph of an object remaining inside the brain. It is a behavioral difference. Describing a retained trace should preserve its status as an inference from the measurement.

What is the percentage actually of?

Suppose, as an instructional example, original learning takes ten minutes and relearning six. The four-minute difference divided by the original ten gives forty percent savings. These are hypothetical values, not experimental observations. Forty percent saved work does not mean forty percent of the material is reproduced exactly item by item.

Ebbinghaus's actual calculations were more detailed. He separated recitation time from learning time and discussed corrections where needed. The simple example is not a replacement for that method. It illustrates a common confusion: a graph labeled retention does not necessarily report the same recall test as every other graph. The source of the percentage has to be identified. [2, §28]

Faster initially, slower later

Ebbinghaus described an initially rapid decline followed by slower change under his conditions. More work had to be repeated in early relearning, while additional differences between longer intervals became smaller. That broad shape is the curve's familiar feature. He nevertheless framed it as a result for a particular individual and particular syllable series. [2, §29]

He also proposed a mathematical approximation, explicitly withholding a claim that it possessed general significance beyond summarizing the observed results. The original account therefore does not supply a universal schedule for all people and materials. A smooth mathematical line does not establish unlimited applicability.

Time of day remained part of the problem

Learning and relearning at different hours could involve different levels of mental freshness. Ebbinghaus discussed the problem of learning in the morning and working again later. Separating a delay's effect from the effect of performing at another time of day was not straightforward. [2, §27]

Elapsed time is not the only thing taking place. Daily life continues during the interval, and conditions change. More observations can help characterize fluctuation without automatically removing every variable. Reporting those difficulties exposes a limit rather than concealing it. Taking a curve seriously is not the same as assuming it has no boundaries.

What a replication added

Murre and Dros's 2015 study applied the savings method again with one participant. They reported results resembling Ebbinghaus's data. The finding supports testing an older result in another investigation, but does not turn the single-participant design into a broad population sample. [3]

Two questions remain distinct: did a similar method produce a similar pattern, and how far does that pattern extend across ages, materials and contexts? Support for the first does not fully answer the second. Studies can strengthen attention to a result without abolishing its limits.

Avoid turning the curve into a verdict about yourself

Lecture notes, faces and work routines need not be assessed with identical material or criteria. Extracting exact repetition times for everyone would go beyond the studies examined here. A more informative beginning is to specify what is being evaluated, after what delay and with what task.

That suggestion is an inference from measurement logic, not an experimental demonstration that a particular study program is superior. Rather than judging ourselves against one historical percentage, we can ask what remembering means in the task. Failing to say something now and starting its relearning entirely from zero can be different. That difference is a central question opened by the investigation.

Evidence-based closing: Testing an unseen effect behaviorally

This article distinguishes Ebbinghaus's method and forgetting chapters from the 2015 replication report. Syllables, learning criteria, intervals and savings calculations are checked against the historical primary source. The ten- and six-minute example is not presented as data. No direct measurement of an entire memory being erased is claimed.

The strongest conclusion is that research need not rely solely on successful conscious recall. Relearning tests a continuing effect of earlier work through another behavioral measure. It does not establish identical losses at identical speeds for everyone. The curve is neither a personal label nor a diagnosis, but a measured answer to a defined experimental question.

Sources

1. Hermann Ebbinghaus, Memory: A Contribution to Experimental Psychology (1885), public-domain translation by Henry A. Ruger and Clara E. Bussenius, 1913, Chapter III: Method, §§11–16.
2. The same work, Chapter VII: Time, forgetting and savings, §§26–30, York University's Classics in the History of Psychology archive.
3. Jaap M. J. Murre and Joeri Dros (2015), Replication and Analysis of Ebbinghaus' Forgetting Curve, PLOS ONE, 10(7), e0120644, DOI: 10.1371/journal.pone.0120644.

The sources were directly inspected on October 14, 2026. This is original explanatory prose. Historical tables and modern research figures have not been reproduced.