HSW-0216 · How Studying Works
You can see the same thing again and again, perform reasonably well each time, and still fail to build much evidence that the pattern has entered long-term memory.
That sounds impossible if repetition is treated as learning itself. Surely the hundredth exposure must leave more behind than the first. Often it does. But a 2025 visual-memory study shows why that intuition needs a boundary: repeated performance can sometimes be supported by a temporary memory system that refreshes the current trial without accumulating a strong long-term trace.
Quick answer
Ye, Liu and Logie repeatedly showed participants the same six-item visual array in a change-detection task. With a short 500-millisecond delay between the study array and the test array, there was no evidence of learning across 120 study trials even though performance stayed well above chance. The authors interpret this as consistent with a limited-capacity visual cache that could hold several items long enough for the current trial, but whose contents were overwritten by the next trial.
With longer study–test intervals, Experiment 1 showed slow improvement after many repetitions for some participants, consistent with gradual support from an episodic long-term memory trace. A second experiment using a new visual array on every trial helped show that ordinary task practice did not explain the first experiment’s slow accuracy gains. The paper also reports important limitations, including modest sample sizes for some mixed-design comparisons.
The studying lesson is not that repetition fails. It is more precise: successful performance during repeated exposure does not prove that the repeated information itself is accumulating in long-term memory.
Performance can be carried by the moment
Imagine looking at six coloured shapes arranged around a circle. A moment later you must decide whether one feature changed. You can succeed by holding enough of the current array temporarily. You do not necessarily need a durable record that survives into tomorrow, or even into the next trial.
This distinction is fundamental. A task can be completed using information that is currently active without requiring that information to become a stable long-term memory.
Students encounter the same problem whenever the material remains visible. They can compare a worked example with the next question, reread a definition while answering, or recognise a diagram that is still on the screen. Performance is real, but the source of performance may be the environment plus temporary memory rather than independent learning.
What the visual-memory experiments did
In Experiment 1, participants repeatedly studied the same array of six objects, each defined by shape, colour and location. They then completed a change-detection test. Different groups experienced study–test intervals of 500, 2,000 or 5,000 milliseconds.
At the shortest interval, accuracy remained around a relatively stable level across the full series instead of steadily rising with repeated exposure. The same array had been presented again and again, but participants behaved as though each trial was largely being solved afresh.
With longer intervals, the first experiment showed slower, delayed improvement: group performance began improving only after many repetitions, and some participants never showed strong evidence of learning the repeated array. Participants who became aware that the array was repeating were more likely to show improvement.
Experiment 2 replaced the repeated array with a new array on every trial. Accuracy did not show the same slow repeated-array learning pattern, helping the authors distinguish learning of the repeated stimulus from simple practice at pressing the right response button or getting used to the task.
The visual-cache interpretation
The authors interpret the results within a framework where a temporary visual cache can support several visual feature bindings for the immediate task. On this account, the cache is useful but limited. Once the next display arrives, its contents can be overwritten—even when the next display happens to be identical.
That explains an otherwise puzzling pattern: a participant can perform at 70–80% across many trials without showing the upward curve expected if a strong long-term representation of the entire six-item array were steadily accumulating.
The paper contrasts this interpretation with alternatives in which working memory consists of activated long-term-memory traces. The data do not close the theoretical debate. They provide evidence that short-term task support and long-term learning can behave differently enough that one cannot be inferred automatically from the other.
Why repetition usually works better than this sounds
The same paper notes that many classic repetition studies do show learning, often much faster, especially for recall of repeated sequences. The unusual result concerns a particular visual change-detection setting where a temporary representation can solve much of the task.
This matters because repetition is not one operation. Repeating can involve:
- re-seeing;
- re-reading;
- reconstructing from memory;
- producing a response;
- comparing alternatives;
- noticing an error;
- explaining a relation;
- performing a sequence;
- retrieving after delay.
Two study sessions can both contain “ten repetitions” while demanding completely different cognitive work.
Worked example: the formula beside the question
Illustrative case. A student completes twenty physics questions while a formula sheet remains open beside the notebook. The student gets faster and more accurate.
That improvement may include genuine learning. But it may also include faster visual search, better recognition of which line to use, and temporary retention of the formula from one question to the next. The visible sheet is carrying part of the performance.
Close the sheet tomorrow and ask for the formula, its conditions and one application. Only then do you know how much of yesterday’s success travelled with the learner.
Worked example: diagram labelling
Illustrative case. A student labels the same cell diagram five times from worksheets where faint guide lines and familiar spatial positions remain constant. Accuracy reaches 100%.
Then the diagram is rotated, simplified and presented without arrows. Performance collapses. The learner may have improved at reading the repeated display rather than building a flexible representation of structure.
The solution is not endless extra copies of the same diagram. It is to change what the repetition demands: recall labels before viewing, redraw from memory, explain function, rotate the representation and identify the same structure again.
The exposure–reconstruction distinction
Exposure puts the information in front of the learner again. Reconstruction asks the learner to rebuild something after some of the support has disappeared.
Both have uses. Exposure can establish familiarity, provide examples and correct misconceptions. Reconstruction tests whether the learner can generate the representation rather than merely use a representation that remains available.
If every repetition leaves the original pattern visible, the study design may never require the memory system to demonstrate that the pattern survived.
Do not mistake faster responding for stimulus learning
In the visual-memory paper, response times became faster across trials even in conditions where accuracy did not show evidence of learning the repeated visual pattern. That is an important diagnostic warning.
People can become faster because they learn the task routine: where to look, how to respond, what button to press, what counts as a change. Speed improvement may reflect procedural familiarity with the testing environment rather than stronger memory for the target information.
Students show the same pattern when they become fast at a worksheet format. The format has been learned. Whether the underlying knowledge has been learned is a separate question.
The support-removal test
After several successful repetitions, remove one support and see what happens.
- Hide the example.
- Delay the next attempt.
- Change the layout.
- Remove the labels.
- Ask for production instead of recognition.
- Change the cue while preserving the target knowledge.
If performance remains strong, the learner has better evidence that the capability is not being carried entirely by the immediate display.
When awareness matters
In the repeated-array work, participants who became aware of the repetition were more likely to show learning. The researchers discuss awareness as related to the strengthening of an episodic trace rather than treating it as a simple switch that causes memory.
For studying, awareness can still be a useful practical lever. A learner who is told, “This same relationship is appearing again—what is invariant?” may stop treating each item as a fresh display and begin encoding the recurring structure intentionally.
That is a teaching application, not a direct result of the experiment. It should be tested through later performance rather than assumed to work because the analogy sounds plausible.
Repetition should change state
A useful study progression makes repeated contact with the same knowledge increasingly independent.
- See it: understand the model or example.
- Complete it: fill missing parts with support.
- Reconstruct it: reproduce from reduced cues.
- Retrieve it: produce after delay.
- Use it: apply in a changed problem.
- Explain it: justify why the relation holds.
The knowledge is repeated, but the task is not. Each pass asks the learner to carry more of the performance internally.
Working memory is not failure
Temporary visual support is not a defect. Working memory exists because the mind needs to keep current information available long enough to act on it. A student reading a graph, following a geometric construction or comparing two sentences must hold information transiently.
The mistake is to infer long-term learning merely because working memory successfully supported immediate performance. For the broader function of holding information long enough to think, see How Intelligence Works | Working Memory.
A delayed independent check
After the repeated practice ends, wait long enough for the immediate display and its transient support to disappear. Then test the same underlying knowledge with no original model visible.
Use at least two checks:
- Reconstruction: can the learner reproduce the essential structure?
- Transfer: can the learner recognise or use the same structure when surface details change?
If yesterday’s repeated performance does not survive either test, the original practice needs a different learning demand, not merely more identical exposures.
For parents and tutors
When a learner says, “I did this ten times already,” do not argue about the number. Ask what the ten repetitions required. Was the answer visible? Was the diagram unchanged? Did the same cue start every problem? Did the student have to reconstruct anything after a delay?
If repetitions were mostly exposure, change the next repetition rather than simply adding an eleventh copy. Remove support, delay retrieval, change representation or ask for explanation.
What not to conclude
The Ye, Liu and Logie experiments do not show that repeated visual study is generally useless. The task involved abstract six-item visual arrays and change detection. Many forms of repeated sequence learning show clear improvement, and the paper itself contrasts the present findings with faster repetition learning found in other paradigms.
The paper also acknowledges statistical limitations, including modest participant numbers in a mixed design. Its strongest contribution is the converging pattern: under short study–test intervals, many repeated presentations produced stable immediate performance without evidence of progressive learning of the entire repeated array.
The deeper studying principle
Repetition counts encounters. Learning asks what changed because of them.
If the same support remains available on every attempt, the learner may become excellent at using the support. That can be useful. But when the educational goal is independent capability, the support must eventually disappear long enough for learning to reveal itself.
Do not ask only, “How many times have I seen this?” Ask, “What can I now reconstruct when it is gone?”
Research basis and routes
Primary research: Ye, C., Liu, Q. & Logie, R. H. (2026 issue; published online 29 November 2025), “When repeated presentation of visual feature bindings does and does not result in learning: Visual short-term and long-term memory are distinct but work in tandem”, Memory & Cognition. The authors provide data and code through the Open Science Framework and explicitly discuss alternative theoretical accounts and power limitations.
For the broader memory workspace, see How Intelligence Works | Working Memory. For study material that disappears before it can be integrated, see How Transient Information Works.
Continue through the How Studying Works Numbered Series Reading Index or return to the How X Works Hub.