Interleaving: the counterintuitive ordering effect.

Of all the scheduling levers a course designer controls, the humblest is the order of the practice problems. That decision is usually left to whoever assembled the worksheet, on the theory that it cannot matter much. The interleaving literature exists because that theory was tested. The order turned out to move delayed-test scores more than most content revisions ever do.

Open almost any textbook, course, or training module and you will find the same shape: a lesson on one technique, then a set of exercises that all use that technique. Practice one thing until it feels smooth, then move to the next. This is blocked practice. It is how nearly all instruction is organized — because it feels efficient to learners and looks efficient to teachers.

The alternative sounds almost careless. Shuffle problems from several related topics together, so that back-to-back items rarely call for the same approach. Learners under this interleaved schedule make more errors during practice, report the sessions as harder, and rate the method as less effective. Then they sit a delayed test — and they win, often by a wide margin. That reversal between what practice feels like and what it produces is one of the most instructive findings in learning science. It explains why so much well-meant curriculum design is organized backwards.

A shuffled problem set, a reversed result

The cleanest demonstration comes from a study in which college students learned to compute the volumes of four obscure solids (Rohrer & Taylor, 2007). Every student solved the same practice problems; the only thing varied was order. One group worked in blocks — all problems of one type, then all of the next. The other worked through the same problems shuffled together.

During practice, the blocked group was far more accurate. No surprise: each problem repeated the procedure just used, so there was nothing to figure out. One week later, on a test of new problems, the ordering reversed sharply. The interleaved group answered 63% correctly, against 20% for the blocked group (Rohrer & Taylor, 2007).

The reading the authors offered has held up well since. Under blocking, a problem announces its own solution: if the last ten problems were wedge problems, the eleventh is too. Students practice doing a procedure but never practice picking one. A test — like every real situation in which knowledge is used — does not announce which technique applies. Interleaved practice is practice at choosing. Blocked practice is practice at carrying out a choice someone else already made.

Twelve painters and a stubborn illusion

A single mathematics result, however striking, could be a quirk of procedures. Perhaps mixing only helps when the skill is choosing among formulas. The literature’s next landmark closed that escape route by moving to a domain with no formulas at all. It also added the finding that made interleaving famous beyond scheduling: the learners’ own judgment of what was helping them was exactly backwards.

Interleaving is not just for procedures. A now-classic study looked at inductive learning — learning a category from examples. People studied landscape paintings by twelve artists, either blocked — several works by one artist in a row — or interleaved, with the artists shuffled together (Kornell & Bjork, 2008). The test was to name the artists of paintings never seen before.

The expectation, shared by the researchers and by nearly everyone shown the design, was that blocking should help. Seeing an artist’s works side by side ought to let their common style emerge. The data said otherwise. Interleaved study produced reliably better sorting of new paintings. The paper’s title asked whether spacing is the “enemy of induction”; the data answered that it is closer to a friend.

The most important part of the study, though, was metacognitive — it concerned learners’ judgment of their own learning. Asked afterwards which schedule had helped them more, a large majority said blocking — even after their own test scores had shown the opposite (Kornell & Bjork, 2008). The schedule that works loses the exit survey.

The mechanism: learning to tell things apart

Why would a worse-feeling schedule produce better learning? The leading account is discriminative contrast — learning by comparison. Much of what we call “understanding a concept” is really the ability to tell it apart from its look-alike neighbors. This solid from that one; this artist from a similar one; this clause type from the one it resembles. Blocked practice stresses what members of a category share. Interleaving puts categories side by side, so their differences stand out exactly when they can be compared.

Several lines of evidence support this reading. The interleaving advantage in style-learning appears specifically when the schedule promotes contrast between categories — not from spacing alone (Kang & Pashler, 2012). Work isolating the parts suggests two contributors: telling categories apart, and retrieval from long-term memory (Birnbaum, Kornell, Bjork & Bjork, 2013). And the benefit is not unconditional. When categories are easy to tell apart, blocking can compete; interleaving’s advantage concentrates on categories that are genuinely confusable (Carvalho & Goldstone, 2014).

Interleaving also brings spacing along with it, unavoidably — mixing topics spreads each topic’s repetitions out in time. Studies built to separate the two contributions in mathematics find that contrast and spaced practice both do real work (Foster, Mueller, Was, Rawson & Dunlosky, 2019). In practice this is a feature, not a flaw: a well-built interleaved schedule delivers two evidence-backed effects at once.

The meta-analytic picture

The literature has been pooled. A meta-analysis of the interleaving experiments found an overall benefit of interleaved over blocked practice that is small-to-medium on average (Hedges’ g ≈ 0.42) — but heavily dependent on material (Brunmair & Richter, 2019). The benefit is large for confusable visual categories such as paintings, and for mathematics problem types. It shrinks toward zero — in some materials reversing — for content like foreign-language vocabulary, where the items studied do not compete with one another (Brunmair & Richter, 2019).

The meta-analysis is titled “Similarity matters,” and that is the correct summary. Interleaving is not a universal booster; it is a discrimination-training tool. It pays when the learner must tell similar things apart, and it pays most when those things are most easily confused.

From laboratory to classroom

Lab effects earn their place in this library only when they survive real classrooms, real teachers, and delays measured in weeks rather than minutes. That is the graveyard where many polished findings quietly disappear. Interleaving’s classroom record is among the strongest in the collection.

The effect scales beyond the lab. Fourth graders who practiced mixed problem types scored roughly double their blocked-practice peers on a next-day test (Taylor & Rohrer, 2010). With middle-school mathematics students, the interleaving advantage was larger on a test given thirty days after practice than one day after — the gap grows with delay (Rohrer, Dedrick & Stershic, 2015). And a large preregistered randomized controlled trial across dozens of seventh-grade classrooms found a large advantage for interleaved worksheets on a delayed test (d ≈ 0.83) (Rohrer, Dedrick, Hartwig & Cheung, 2020).

Set that against how teaching materials are actually built. Textbooks block by chapter; courses block by module; drill software blocks by skill. The ordering the classroom evidence favors is the one almost no published curriculum uses.

Building the shuffle: what implementation actually involves

Turning the finding into a curriculum is more surgical than “randomize everything.” The meta-analysis’s central moderator — similarity — is the sorting rule (Brunmair & Richter, 2019). The design step is finding the families of confusable material within a syllabus — those families are where interleaving pays, and the only place it should spend its costs. In a compliance curriculum, the confusable family is the set of look-alike clauses that trigger different duties. In a product curriculum, it is the plans whose terms overlap; in a clinical one, the presentations that share symptoms and diverge in treatment. Material outside such families — orientation content, isolated facts, procedures with no neighbors — gains nothing from shuffling and can stay in its natural order.

The second design decision is dosage over time. The classroom trials did not interleave within single sessions only. The strongest designs mixed problem types across assignments spanning weeks, so that every session reopened older families alongside newer ones (Rohrer, Dedrick, Hartwig & Cheung, 2020). That structure is spacing and interleaving delivered as one schedule. It is precisely why hand-built worksheets struggle to sustain it, and why the technique historically stayed in labs. Keeping a rolling, per-learner mix of confusable items across a term is bookkeeping — and bookkeeping is what software is for.

The third decision is what to tell learners, and the self-judgment results make it non-optional. Interleaved practice feels worse while working better (Kornell & Bjork, 2008), (Yan, Bjork & Bjork, 2016). So an unexplained interleaved course reads as a disorganized one — and learners who conclude the course is broken pull back from the very difficulty that is teaching them. The fix costs a paragraph. Tell learners the mixing is deliberate, name the effect, and set the expectation that practice will feel harder and test better. Instructors need the same briefing for the same reason; the schedule that works looks untidy from the podium too.

Why blocked practice fools learners and instructors

The company version of the illusion deserves naming alongside the individual one, because buying decisions run on it. Courses are piloted, demoed, and renewed on within-session impressions: smooth progress, satisfied learners, high end-of-module scores. That systematically favors blocked designs, for exactly the reasons the lab documents. A vendor whose product interleaves honestly will demo worse than one whose product blocks, in front of every buyer who judges by watching a session. The countermeasure is the same one the researchers use: judge on delayed performance, never on the experience of the session itself.

The deeper lesson is about measurement. Performance during learning is a poor stand-in for learning — and in this literature it points the wrong way. Under blocking, practice is fast, fluent, and light on errors. Learners read that fluency as mastery. Instructors watching the same session see smooth progress and satisfied students, and draw the same wrong conclusion.

The illusion is remarkably hard to correct. Even when people experience both schedules and see their own interleaved scores come out ahead, most continue to believe blocking worked better. They credit the smoothness of blocked study to better learning (Yan, Bjork & Bjork, 2016). Both sides of the classroom, in other words, prefer the schedule that produces less durable learning. That is why the correction has to be built into the system that sequences practice, not left to judgment in the moment.

What the evidence doesn’t show

Interleaving has earned its place in the evidence-based toolkit. But the literature is specific about its boundaries, and honest applications should be too.

• It is not “mix anything with anything.” The benefit depends on the interleaved items being confusable. Alternating algebra with history trains no discrimination, and meta-analytic estimates for low-similarity materials such as foreign vocabulary are near zero or negative (Brunmair & Richter, 2019).
• Blocking is not always wrong. When categories are easy to distinguish, blocked study can match or beat interleaving (Carvalho & Goldstone, 2014), and hybrid schedules — a short blocked introduction before mixing — remain under-studied relative to how often they are recommended.
• The long-horizon evidence is concentrated in mathematics. The strongest classroom trials involve school math over delays of days to a month (Rohrer, Dedrick, Hartwig & Cheung, 2020). Extension to workplace skills and multi-month retention is plausible but rests on a thinner base.
• Interleaving and spacing travel together. Many designs cannot fully separate the two, and where they are separated, both mechanisms contribute (Foster et al., 2019) — so “interleaving” gains reported in applied settings are usually a package, not a single ingredient.

Where the evidence stops

1. 1It is not “mix anything with anything.”
2. 2Blocking is not always wrong
3. 3The long-horizon evidence is concentrated in mathematics
4. 4Interleaving and spacing travel together

© 2026 FUTURE PROOF™

What this means for practice

Interleaving also pairs with retrieval in a way worth making explicit. A well-built interleaved session is almost unavoidably a testing session. Every item opens with the discrimination question — which kind is this? — and only retrieval can answer it. The practical result: the two strongest scheduling effects in this library arrive as one design. Shuffle the confusable families and deliver them as questions, and the session is at once interleaved, spaced, and retrieval-based — no further engineering needed (Foster et al., 2019).

Treat sequence as a design variable with the same status as content. Shuffle practice within families of confusable problem types, not across unrelated subjects. Expect — and tell learners to expect — lower accuracy during practice; the dip is the mechanism working, not a defect. Judge on delayed tests rather than end-of-session scores. Session performance rewards exactly the wrong schedule, and it will keep doing so for as long as anyone watches it. And treat a learner’s specific confusions as a map: the pairs of ideas a person actually mixes up are the highest-value candidates for interleaved contrast.