When Should You Study an Example Instead of Testing Yourself?

Suppose you are learning a new kind of math problem. You have read the explanation once, close the book, and try to solve one from scratch because retrieval practice is supposed to beat rereading. Five minutes later you are still staring at the first line.

That does not necessarily mean the difficulty is helping you learn. You may simply not have a usable procedure to retrieve yet.

Worked examples and self-testing do different jobs. A worked example can show you what a successful solution looks like and how its parts fit together. Retrieval or independent problem solving makes you produce knowledge without having the answer in front of you. Both are useful. The practical question is which failure you are trying to fix.

If you do not yet see how the solution works, study the solution

In a 2026 study, Meng Cao and Paulo Carvalho compared worked examples with a relatively constrained form of retrieval practice: learners selected the correct rule or method from four options and received immediate feedback. When learners first received explicit instruction in the underlying rules, retrieval practice produced better generalization to new problems than studying worked examples. When that instruction was removed and learners had to infer the rules while they practiced, worked examples produced better generalization than repeated retrieval items. Retrieval caught up when the practice items were varied. Cao and Carvalho, 2026

That experiment was not a direct test of someone solving a blank-page problem from scratch. Its useful contribution is narrower: it isolates a difference between practicing a rule that has already been supplied and having to infer the rule in the first place.

That pattern suggests a useful way to think about the first stage of learning a procedure. If you cannot yet explain what a competent solution is doing, a blank practice problem may demand too much search. A clear worked example can give you a structure to inspect.

That does not mean reading the example from top to bottom and saying, “Yes, that makes sense.” Recognition is cheap. Try to explain why each consequential step is there. What changed between this line and the next one? What rule allowed that move? What would go wrong if you skipped it?

For a difficult procedure, one good example may do more useful work than several failed attempts made before you know what to attempt.

Once you understand it, start taking the example away

The advantage of examples does not mean you should keep studying them indefinitely.

A 2025 classroom experiment with 166 Dutch fifth graders looked specifically at what happened after initial acquisition. Students practiced by either studying examples or solving problems. Five minutes later, the groups did not differ significantly in problem-solving performance. One week later, the students who had practiced by solving problems performed better. The result held across the study's simple and more complex task conditions. Ruitenburg et al., 2025

This is the point at which a worked example can become a comfort blanket. The learner can follow every step and mistake that fluency for an ability to produce the steps later.

A better transition is gradual. After you can explain an example, cover part of it. Try to supply the next step. Then check. Move to a partially completed problem. Eventually remove the scaffolding and solve the whole thing.

There is good evidence for this middle ground. In a 2025 experiment, Gesa van den Broek, Margot van Wermeskerken, and Tamara van Gog gave adults step-by-step worked examples. Some participants simply revealed each successive step. Others had to say what the next step should be before it appeared. One week later, the retrieval-prompt group performed better both at recalling the procedure and at solving problems. van den Broek, van Wermeskerken, and van Gog, 2025

The improvement was not free. The prompted group spent about 31 minutes studying, compared with about 13 minutes for the control group. That matters if you are deciding how to spend a limited study session. Still, it shows that example study and retrieval practice do not have to be competing camps. An example can provide the structure while retrieval gradually forces you to own more of it.

A simple version requires no special software: look at the first step, hide the next one, predict it, then reveal it.

Knowing the rule and using the rule are not the same problem

There is another reason not to turn self-testing into a universal answer. Sometimes the thing you are testing is not the thing you are failing to do.

A 2026 classroom study by Fieke H. A. Ophuis-Cox, Leen Catrysse, and Gino Camp taught Dutch fourth graders verb-spelling rules using either worked examples alone or worked examples plus retrieval practice. A week later, the combined group remembered the rules better. But it did not significantly outperform the worked-example-only group at applying those rules correctly. Ophuis-Cox, Catrysse, and Camp, 2026

That distinction is easy to miss in ordinary studying. You can correctly state the rule for choosing a statistical test and still choose the wrong one when given a new study design. You can memorize the steps of a programming pattern and fail to notice when a problem calls for it. You can know a grammar rule and still misapply it in a sentence.

If recall is weak, test recall. If application is weak, practice application.

That sounds obvious once stated, but many study systems measure the easiest thing to quiz. A flashcard can tell you whether you remember a definition. It cannot by itself tell you whether you will recognize the situation in which the definition becomes useful.

Use examples when the bottleneck is understanding; use testing when the bottleneck is production

For most procedural learning, the following progression is more useful than choosing one favored technique and applying it everywhere.

Start with a worked example when the procedure is still opaque. Choose a representative example that exposes the important steps clearly. Early examples should not be puzzles in their own right.

Explain the example rather than merely rereading it. Ask why the steps work and what feature of the problem made each one appropriate. This helps turn the visible solution into a model you can use elsewhere.

Begin retrieving pieces of the procedure. Hide the next step. Complete a missing line. Predict what happens next before looking. Failed retrieval should be followed by the answer, not rehearsed indefinitely.

Move to whole problems when you can reconstruct the procedure. At this point, independent problem solving is doing work that another example cannot: forcing you to choose and produce the method yourself.

Vary the problems once the basic method is stable. Transfer requires more than reproducing the same solution with different numbers. You need to learn which features matter, which do not, and when the procedure applies at all.

Come back later. A procedure that works five minutes after studying can still disappear by next week. Delayed retrieval and problem solving are a different test from immediate fluency.

This sequence is not a law. Some domains are less structured than algebra or spelling, and worked examples can remain useful for experienced learners when the interesting work lies in judgment rather than a fixed procedure. Prior knowledge also changes what counts as an example. To a novice, a proof may be a model to imitate; to an expert, the same proof may be a case to compare with several alternatives.

The useful question is narrower: what would make the next round of study productive?

If you are looking at a solution and still cannot explain how it works, another blank test may only confirm that you cannot yet produce it. Study the example more carefully, or find a clearer one. Once the structure is intelligible, begin removing it. The point of the example was to make independent performance possible, not to become a permanent substitute for it.

Sources

• Meng Cao and Paulo F. Carvalho, “Striking the Balance: How Variability Shapes Retrieval Practice and Worked Examples for Transfer Learning,” Educational Psychology Review 38 (2026), article 71
• Gesa S. E. van den Broek, Margot van Wermeskerken, and Tamara van Gog, “Retrieval Practice in Stepwise Worked Examples Improves Learning,” Learning and Instruction 100 (2025), 102196
• Sterre K. Ruitenburg et al., “After initial acquisition, problem-solving leads to better long-term performance than example study, even for complex tasks,” Learning and Instruction 95 (2025), 102027
• Fieke H. A. Ophuis-Cox, Leen Catrysse, and Gino Camp, “Mixed effects of retrieval practice on the acquisition of procedural knowledge in verb spelling education,” Learning and Instruction 104 (2026), 102377