Why slowing down helps
When Robert Bjork introduced the term “desirable difficulties” in the early 1990s, he was naming a counterintuitive pattern that had been accumulating in the literature for decades: the manipulations that speed up visible progress during practice tend to produce fragile learning, while the ones that impede it often produce durable memory and better transfer.
The core observation is a dissociation between acquisition and retention. Massed practice — working through the same material repeatedly in a single session — produces fast, smooth performance that falls apart on a delayed test. Spaced practice slows the session down, introducing forgetting that has to be overcome, and that effortful recovery is precisely what consolidates the memory. The same logic runs through interleaving, retrieval practice, and reduced feedback: each one makes the task feel harder in the moment and performs better a week later.
The word desirable is doing real work here. Not every difficulty helps. Bjork’s framework distinguishes conditions that tax working memory without building anything — poor font choice, irrelevant complexity, a badly sequenced explanation — from conditions that engage the processes actually responsible for long-term storage and flexible use.
Testing yourself instead of rereading, spacing study sessions so that some forgetting has occurred before review, and mixing problem types rather than blocking by category all qualify.
What qualifies and what doesn’t
The difficulties that earn the label share a common mechanism: they force retrieval or reconstruction rather than passive re-exposure. Each intervention has a body of experimental support, and most of that support holds reasonably well outside the laboratory, which is not something you can say about every effect in the cognitive-science literature.
The evidence is not uniform, however. The spacing effect replicates robustly across materials and populations; interleaving shows clear gains in mathematics and category learning but the data are thinner elsewhere. Expanding retrieval intervals — the idea behind spaced-repetition software — have a more contested record than the software’s marketing implies; fixed intervals sometimes perform just as well.
Elizabeth Bjork’s work on contextual interference has clarified when interleaving helps most: the benefit is largest when the learner already has some command of each item or skill being shuffled, so that switching demands discrimination rather than simply inducing confusion.
Judgements of learning are the psychological mechanism that makes desirable difficulties hard to adopt in practice. Learners — and the people who design courses for them — consistently rate their own mastery by how fluent a session feels. Massed practice feels productive; interleaved practice feels chaotic; re-reading feels efficient; retrieval feels hard. These subjective ratings are inversely related to actual retention on a delayed test, which is why well-intentioned people reach for precisely the wrong tools. The difficulty is desirable; the feeling it produces is not.
How robust is the framework?
“Desirable difficulties” is a framework rather than a single effect, and that breadth is both its strength and its weakness. As a unifying account it is elegant: it explains why so many effective interventions feel bad and why so many comfortable ones fail. But it resists clean falsification, because any difficulty that fails to improve retention can be reclassified as undesirable. Critics have pointed out that the line between the two is sometimes drawn after the fact, making the framework unfalsifiable in practice even if productive in research.
What survives the strongest scrutiny are the specific instantiations. Spacing, retrieval practice, and interleaving each have effect sizes large enough to matter, laboratory-to-real-world records solid enough to take seriously, and replication histories that have held up better than much of cognitive psychology’s catalogue. The overarching label is a heuristic; the individual effects are the evidence.