A typical practice set in a mathematics textbook follows a familiar pattern. A lesson introduces a new procedure, followed by twenty problems that all require that procedure. Students work through them, accuracy climbs, and both student and teacher reasonably conclude that learning has occurred. Then the unit test arrives, problem types are mixed together, and performance drops.
This gap between performance during practice and performance later is one of the most consistent findings in learning research. It suggests that practice problems do more than provide repetitions. They shape what students attend to, which cues they rely on, and what kind of knowledge they build. A problem set can strengthen a student's ability to execute a procedure while leaving untouched the more important ability to recognise when that procedure applies.
Memory science offers a useful lens for this problem. The features of individual problems, the order in which they appear, and the variation across a set all influence whether practice produces durable, transferable understanding or a temporary sense of fluency. Each of these can be designed deliberately, and the research gives clear guidance on how.
Problem Features: What a Problem Asks the Learner to Think About
Every practice problem has two layers. Surface features include the context, the numbers, and the wording. Structural features are the underlying principle that determines how the problem should be solved. In a well-known study by Chi, Feltovich and Glaser, novice physics students sorted problems by surface features such as inclined planes or pulleys, while experts sorted them by principles such as conservation of energy. Expertise, in large part, is the ability to see structure beneath surface.
Many practice problems quietly undermine this ability. When every problem on a page follows a lesson on the same method, the context gives the answer to the most important question: which approach should I use? Tulving's work on encoding specificity is relevant here. Memories are retrieved most readily when the cues present at retrieval match those present at learning. If the main cue during practice is 'this is the chapter on ratios', students build knowledge tied to that episode rather than general, semantic understanding they can apply elsewhere.
A second important feature is the degree of effort a problem demands. Problems that require learners to retrieve information and generate solutions produce stronger memory than those that allow copying a nearby model. Bjork describes these as desirable difficulties. However, difficulty is only desirable when the learner can meet it. For novices, cognitive load research by Sweller and colleagues shows that studying worked examples often produces better learning than solving problems unaided, because unguided problem solving can consume working memory without building useful schemas.
Taken together, effective problems tend to share several characteristics. They require retrieval rather than recognition. They demand that learners select a strategy, not just apply one. They vary in surface features while sharing deep structure, or look similar while requiring different approaches. And they are pitched at a level of challenge the learner can productively engage with, given the support available.
TakeawayA practice problem teaches whatever it makes the learner think about. If the context gives away the method, the problem practises execution but not recognition.
Sequence Matters: The Order of Problems Shapes What Is Learned
The most studied question in problem sequencing is whether to block problems by type or interleave them. In blocked practice, students complete all problems of one kind before moving to the next. In interleaved practice, different types are mixed. In a classroom study by Rohrer, Dedrick and Stershic, seventh-grade students who practised with interleaved problems scored 72 percent on a test one month later, compared with 38 percent for those who practised in blocks. A later large randomised trial across dozens of classrooms produced similarly strong effects.
The explanation lies in what each sequence requires. Blocked practice lets students carry one strategy from problem to problem without deciding whether it fits. Interleaved practice forces a discrimination at every step: what kind of problem is this, and what does it call for? That decision is exactly what tests and real tasks demand. Interestingly, blocked practice usually produces higher accuracy during practice itself, and learners consistently judge it more effective. Kornell and Bjork found that most participants believed blocking helped them more, even when their own results showed the opposite.
Spacing is closely related. Distributing practice on a topic across days and weeks, rather than concentrating it in a single session, is among the most robust effects in memory research. Hattie's syntheses of meta-analyses place spaced practice among the higher-impact influences on achievement. Interleaving and spacing often work together, since mixing problem types naturally spreads practice of each type across time.
Sequence also matters within the learning of a single skill. Renkl and Atkinson showed the value of faded worked examples: beginning with fully worked solutions, then offering partially completed problems, then independent ones. This progression respects the expertise reversal effect described by Kalyuga, in which supports that help novices become redundant and even counterproductive for more advanced learners. Good sequencing tracks the learner's growing competence.
TakeawayPerformance during practice and learning from practice are different measurements. Sequences that feel smooth often produce weaker retention than those that feel effortful.
Design Principles: Building Practice Sets That Promote Deep Learning
The research points to a practical structure rather than a single rule. When a procedure is first introduced, a short run of similar problems is reasonable, because learners need to establish the basic steps before they can discriminate between methods. Pairing these early problems with worked examples reduces unproductive struggle. Once the procedure is functional, practice should shift quickly toward mixed sets.
The second principle is cumulative review. Each assignment can include problems from earlier units alongside current material. A set of twelve problems might include four on this week's topic and eight drawn from previous weeks. This builds spacing and interleaving into routine practice without requiring new materials. A third principle is deliberate variation: change contexts and wording while holding structure constant, and include problems that look alike but require different solutions. Such contrasts direct attention to the features that actually matter.
Fourth, make reasoning visible. Prompts such as 'Which strategy fits this problem, and why?' encourage self-explanation, which Chi and colleagues found strongly associated with deeper learning. Some teachers ask students to classify a set of problems by solution type without solving them, isolating the recognition skill that standard practice neglects. Fifth, provide timely feedback, so that errors are corrected before they are rehearsed and consolidated.
Finally, evaluate practice by the right measure. High accuracy during practice is weak evidence of learning. A more informative test is delayed, mixed performance: can students solve these problems weeks later, when types are combined and no chapter heading signals the method? This also suggests that quantity is a poor proxy for quality. Ten carefully varied problems that require thought often accomplish more than thirty that require only repetition.
TakeawayAsk of any problem set what students will have to decide, not just what they will have to do. The decisions are where transferable understanding is built.
Practice is one of the most powerful tools available to educators, but its effects depend heavily on design. Repetition alone builds a fluency that often fails to survive a change of context or the passage of time.
The evidence supports several clear recommendations. Use faded worked examples when learners are new to a skill. Interleave problem types once basic procedures are established. Build cumulative review into regular assignments. Vary surface features to highlight underlying structure, and ask students to select and justify strategies.
Most importantly, judge practice by delayed, mixed performance rather than immediate accuracy. The goal is not students who can solve twenty identical problems today, but students who can recognise and solve the right problem next month.