Interleaving

type: technique · origin: study · evidence: supported · themes: learning and memory

Claim assessed: Mixing problem types during practice leads to better test results than working in blocks.

Published 2 October 2026, updated 7 October 2026 · how we work

In short

Interleaving means mixing the types of problems when you practise, instead of doing all the problems of one kind first and then those of another. It feels harder, and you even do worse while practising, but results on later tests are better. A meta-analysis found a moderate effect, and a large study in 54 classes found a large effect in maths. The effect does, however, depend on what you are learning.

What it says

Blocks or a mix. Most maths assignments are blocked: a lesson on slope is followed by a dozen slope problems. In an interleaved assignment, no two consecutive problems are solved with the same method. Rohrer et al. (2020) cite an analysis of six US seventh-grade textbooks: on average, 78% of the problems were blocked and only 11% interleaved.

Why it might help. Rohrer et al. (2020) give three reasons:

• When the problems are mixed, you have to choose the method yourself, based on the problem, just as in a test. When they are blocked, you know the method before you read the problem.
• Interleaving spreads the practice of each type of problem over time.
• To solve a problem of a different type from the previous one, you have to recall the formula rather than copy it from the problem before.

The last two are the very techniques in the entry on spacing and retrieval practice.

The first studies. Rohrer and Taylor (2007) had college students practise volume problems in two sessions a week apart. Those who practised interleaved scored worse during practice, but better on the test a week later. Kornell and Bjork (2008) showed participants paintings by several artists, either grouped by artist or mixed. Those who saw them mixed were better at recognising each artist’s style in new paintings. Yet most participants believed grouping by artist had helped them more, even after their own test had shown the opposite.

The large study. Rohrer, Dedrick, Hartwig and Cheung (2020) ran a preregistered randomised study in 54 seventh-grade classes at five schools in Florida, with 15 teachers. Each teacher had classes in both groups. For four months, the classes got eight maths worksheets, with the problems either interleaved or blocked. Then all classes got the same interleaved review worksheet. A month after the review, the students took a surprise test with new problems. Those who had practised interleaved solved 61% of the problems on average, against 38% (d = 0.83). The effect appeared for all 15 teachers. The teachers used the worksheets without any training, and most said afterwards that they would recommend them to others.

The meta-analysis. Brunmair and Richter (2019) gathered 59 studies, with 238 effects. The overall effect was moderate (g = 0.42), but it depended a lot on the material:

• largest for paintings (0.67) and other visual materials;
• small for maths problems (0.34);
• unclear for expository texts and tastes;
• reversed for words, where blocks worked better (−0.39).

Interleaving helped more when the categories resembled each other, when the examples within a category differed, and when the material was complex.

Example

You are preparing for a statistics exam. Instead of solving ten problems on the mean, then ten on probability, then ten on correlation, you make a list in which the types take turns: a probability problem, a correlation one, a mean one and so on. For each one, you first have to ask yourself what kind of problem it is. The example is an illustration, not a study cited here.

How to apply it

The steps below are a practical way of applying it that we propose, based on the cited studies.

1. First learn each type on its own. Rohrer et al. (2020) note that in almost all the maths studies, students had already done a little blocked practice before the mix.
2. Then mix types that can be confused. The effect is larger when the categories resemble each other. What matters is choosing the method yourself, not just switching subjects.
3. Check your answers. In all the maths studies cited by Rohrer et al., students got the solutions and corrected their mistakes.
4. Don’t go by how hard it feels. Interleaved practice feels slower and produces more mistakes in the moment. That doesn’t mean you are learning less.
5. For word lists, don’t insist. In the meta-analysis, blocks worked better for words.

Limits and nuances

• The effect depends on the material. It is clear for maths and visual categories, unclear for texts and reversed for words.
• The large study comes from the team that has studied the idea most in maths. Doug Rohrer is a co-author of four of the ten earlier maths studies reviewed in the article.
• Interleaved practice takes longer. All the teachers in the large study said the interleaved worksheets took more time. Time was not measured, so we don’t know how much of the effect is due to the extra time.
• The effect may be smaller on tests given soon after. Rohrer et al. (2020) note that the only study with no effect gave its test after 2–5 days, and that in another study the effect was smaller on a sooner test.
• The 61% against 38% gap comes from a single study. The meta-analysis finds a smaller average effect for maths (g = 0.34).
• A single large study. The claim rests on a single large, preregistered study, Rohrer et al. (2020), and on a meta-analysis we know only from its abstract. We don’t know whether the meta-analytic effect survives correction for unpublished studies. That is why the level is “supported”, not “strong”.
• What we read. Rohrer et al. (2020) we read in full, in the authors’ version on ERIC. For Kornell and Bjork (2008) and Brunmair and Richter (2019) we read the abstracts; the full text of the meta-analysis is not open access. Rohrer and Taylor (2007) we describe through Rohrer et al. (2020).
• Revised on 7 October 2026. We moved the evidence level from “strong” to “supported”: “strong” requires large replications or a meta-analysis that survives correction for unpublished studies, and here we have a single large study. The content has not changed.

Sources

• Doug Rohrer, Kelli Taylor (2007). The shuffling of mathematics problems improves learning
• Nate Kornell, Robert A. Bjork (2008). Learning concepts and categories: Is spacing the “enemy of induction”?
• Matthias Brunmair, Tobias Richter (2019). Similarity matters: A meta-analysis of interleaved learning and its moderators
• Doug Rohrer, Robert F. Dedrick, Marissa K. Hartwig, Chi-Ngai Cheung (2020). A randomized controlled trial of interleaved mathematics practice

See also: Learning styles, Spacing and retrieval practice