A student rereads a page until it feels familiar. Another closes the book and tries to reconstruct the idea from memory, making mistakes before checking the answer. In the moment, the rereader may feel more successful. A later test can tell a different story.
Desirable difficulties are learning conditions that can make practice feel slower or harder while improving later retention or transfer. The phrase is associated with Robert Bjork’s work and later writing with Elizabeth Bjork. It is not a license to make learning unpleasant. A difficulty is desirable only when it helps the learner build usable knowledge under the conditions that matter.
Learning is not the same as today’s performance
During practice, a learner can perform well because the answer is still visible, the problem type is predictable or the same movement has just been repeated. That smooth performance may reflect temporary accessibility. Long-term learning is tested after a delay, or in a new situation where the cues are less familiar.
Soderstrom and Bjork’s review distinguishes learning from performance and explains why the two can move in different directions. A manipulation that makes today’s practice less fluent may support later retrieval. Conversely, a manipulation that produces immediate success can leave knowledge fragile. The distinction does not mean immediate performance is worthless. It means it is an incomplete indicator of what will remain.
Robert Bjork’s 1994 discussion placed this problem inside metacognition: learners and instructors need valid ways to judge what has been learned. Feeling familiar with a passage is not the same as being able to explain it after a week. The useful response is to test later and under conditions that resemble the intended use.
What makes a difficulty desirable?
The difficulty must engage processes relevant to later learning. Waiting between study sessions can require reconstruction when the material returns. Attempting retrieval can reveal gaps and strengthen future access when feedback follows. Mixing problem types can require the learner to decide which method applies rather than repeating one procedure by pattern.
These examples are different mechanisms, not one magic property called effort. A learner may work very hard deciphering a tiny font and gain nothing. A confusing instruction can consume attention needed for the actual idea. A difficulty that blocks the learner from forming a useful representation is not rescued by being difficult.
Bjork and Bjork’s 2011 account emphasizes creating conditions that benefit long-term learning, not simply adding obstacles. The right challenge depends on what the learner already knows, the material’s complexity, the feedback available and the eventual outcome. The adjective desirable describes the result under those conditions, not how virtuous the struggle feels.
Spacing and the return after forgetting
One well-studied example is spacing practice over time. A gap makes the next encounter less fluent than an immediate repeat. That may invite more reconstruction and can improve later retention. Cepeda and colleagues’ quantitative review of verbal-recall studies supports the spacing effect across a range of conditions, while also showing that timing matters.
Spacing is a candidate desirable difficulty because the learner must re-establish access. But a gap that is too long for the material and learner can lead to repeated failure without support. The design question is not “How much can we make them forget?” It is “When should the learner return so the knowledge can be rebuilt and retained for the intended horizon?”
Spacing can be paired with retrieval, but the two should be named separately. A calendar reminder produces a return. A question answered before seeing the solution adds retrieval demand. Feedback helps correct error. A program may use all three, and a study must distinguish their contributions if it claims one caused the outcome.
Interleaving and discrimination
Blocked practice presents many examples of one type before moving to another. Interleaved practice mixes types, asking learners to identify which principle or response applies. This can feel less orderly. In category-learning experiments, Kornell and Bjork examined whether spacing examples of different categories impaired or helped induction. Their results challenged the intuition that keeping all examples of one category together is always best.
The implication is not that every topic should be shuffled immediately. If a novice does not yet know the basic features of a category, mixing too many examples may overload them. The value of interleaving depends on whether discriminating among categories is itself a learning goal and whether the learner has enough foundation to make that discrimination.
For mathematics, practice that mixes problem types can require selecting a method, closer to a later exam or workplace task than a worksheet that announces which procedure to repeat. Yet the schedule still needs worked examples and feedback when the methods are new. Difficulty becomes useful when it targets choice and retrieval, not when it hides essential instruction.
Retrieval as an effortful learning event
Trying to answer before looking can be harder than rereading. The attempt makes the learner confront what is accessible without the page. A successful retrieval can strengthen future access, and an unsuccessful attempt can still guide learning if corrective feedback is timely and understandable. The task must be within reach; repeatedly asking a novice to generate an impossible answer is not a desirable difficulty.
The design can vary. A low-stakes quiz, a spoken explanation, a worked problem or a prompt to recall a concept all place different demands on the learner. The later test should fit the claim. Remembering a label is not the same as applying a principle to a new case. An intervention may improve one and not the other.
For a learner, the metacognitive benefit is as important as the memory benefit. A failed attempt provides evidence about what needs work. It is uncomfortable, but more informative than the feeling of recognizing an answer when it is already on the page. The point is to use that information for correction, not to rank people by how many errors they make during practice.
When difficulty becomes undesirable
Chen, Castro-Alonso, Paas and Sweller reviewed cases where adding difficulty to high-complexity material can hinder learning. Their cognitive-load account emphasizes the interaction between material and prior knowledge. A novice solving a complex problem may need a worked example before independent generation is useful. What is productive for an expert can be overwhelming for someone meeting the elements for the first time.
This is a real theoretical boundary, not an exception to hide. If learning improves only when a learner can respond meaningfully, instructors must adjust challenge to readiness. The result also warns against confusing friction with learning. Making a page harder to read, withholding essential feedback or imposing arbitrary time pressure can increase effort while reducing what is learned.
The same strategy may change status over time. Worked examples may help early acquisition; later, independent retrieval and varied problems may be appropriate. A curriculum can move from support toward challenge as knowledge grows. That progression should be guided by evidence of understanding, not by the assumption that harder is always better.
Evidence and measurement
Imagine two groups learning to identify unfamiliar medical images. One group studies clear examples with explanations. The other is immediately asked to classify difficult cases without having learned the defining features. The second group is working harder, but that work may be spent guessing. If their later classification is worse, the difficulty was not desirable for those learners at that stage. A better comparison might introduce worked examples first and then vary cases once the feature categories have meaning.
Now imagine learners who already know the basic categories. Giving them only easy, blocked examples can make performance look excellent while leaving them unprepared for mixed cases. For them, interleaving or a delayed test may target a real weakness: deciding which category applies without a label announcing it. The same surface difficulty can therefore change value as knowledge develops. This is why prior knowledge belongs in the design, not merely in the background description of participants.
A test of a desirable difficulty must specify the difficulty and compare later outcomes, not merely record that students found a task hard. It should hold relevant factors comparable: time on task, feedback, exposure to content and the outcome measure. If one group receives more study time and more feedback, later success cannot be attributed cleanly to difficulty.
Delayed retention and transfer are especially important. A difficulty can reduce immediate accuracy yet help later recall. That is the pattern the framework predicts in some cases. If only immediate performance is measured, researchers cannot tell whether the slower practice was beneficial or simply slower. If only a very distant test is used, early learning failures may also be missed.
It is useful to ask what the learner experienced and what the learner achieved. Perceived difficulty can influence motivation and willingness to continue. A method with a long-term benefit may be abandoned if it feels punishing and no one explains why. This does not invalidate the cognitive effect; it makes implementation a separate design question.
When the learner judges the method
A student choosing how to study usually sees only immediate feedback. Rereading feels productive because recognition is easy while the answer is visible. Retrieval can feel unproductive because a missed answer is conspicuous. If students use comfort as the only guide, they may abandon a useful method. This is a metacognitive problem, not a failure of motivation.
One response is to make later learning visible. A teacher can compare a short practice check with a delayed check and explain why the results differ. Learners can keep a record of what they could explain without notes after several days, not merely how familiar the page felt on the first evening. Such checks help calibrate judgment. They also reveal when an allegedly desirable difficulty did not help and should be revised.
Feedback must remain humane. An error during practice is information about the current state of learning, not a verdict on ability. A method that helps memory but reliably discourages participation may fail as an educational design. The task is to make productive challenge sustainable through clear goals, appropriate support and a visible route from attempt to correction.
How to use the principle without turning it into a slogan
Begin with the desired outcome: durable recall, discrimination among cases, application to a novel problem or a skill performed after a delay. Give enough initial explanation for the learner to understand the material. Then introduce a challenge connected to that outcome, such as a spaced return, a retrieval prompt or mixed practice. Provide feedback and test later.
If later performance improves, keep asking for whom and under which conditions. A class of beginners, experienced professionals and self-directed learners may need different levels of support. A public-facing learning product should not make its interface confusing merely to claim it has implemented desirable difficulty. The challenge belongs in the learning operation, not in preventable barriers to access.
For the Institute, this is a practice-architecture principle: design opportunities to retrieve, discriminate and apply, while maintaining enough support for people to succeed. That is an institutional synthesis from the research, not an experimentally established recipe for every course or workplace.
Difficulty has to do useful work
A learning activity can feel demanding because it requires retrieval, comparison or discrimination that will matter later. It can also feel demanding because instructions are confusing, a screen is inaccessible or the learner lacks essential prior knowledge. Only the first kind has a plausible route to a desirable learning benefit. The label should be tied to a mechanism and a later outcome, not awarded whenever a lesson is unpleasant.
For example, asking learners to retrieve an answer before showing it can reveal what they know and make feedback meaningful. Asking them to guess indefinitely without feedback may instead reinforce uncertainty. Interleaving examples can help when the learner must learn which approach fits which problem; it may overwhelm someone who has not yet learned the approaches at all. A responsible design varies challenge, observes later performance and adjusts support. Difficulty is an instrument, never the educational goal itself.
What the evidence does not show
The evidence does not show that suffering, confusion or stress are inherently educational. It does not show that every harder task creates stronger memory, or that immediate mistakes are proof of future learning. It does not justify removing explanations from novices or treating inaccessible presentation as a benefit.
Nor does one successful laboratory manipulation prove broad transfer. A practice that improves word recall may not improve complex reasoning. A mixed problem set may help select among known procedures but be too difficult before those procedures are learned. Each claim needs a later outcome matched to the intended capacity.
The book closes
The student who closed the book and tried to explain the idea may have discovered gaps. The student who reread may have enjoyed a smoother hour. Neither feeling settles what was learned. A later, fair test matters. The best difficulty is the one that helps knowledge survive the closed book and become usable when the setting changes.
That is why desirable difficulties are a design question rather than a command to struggle: challenge should make learning more durable, and evidence should decide whether it did.
Research references
- Bjork, R. A. (1994). Memory and metamemory considerations in the training of human beings. In J. Metcalfe & A. P. Shimamura (Eds.), Metacognition: Knowing About Knowing (pp. 185–205). MIT Press. View source
- Bjork, E. L., & Bjork, R. A. (2011). Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning. In M. A. Gernsbacher et al. (Eds.), Psychology and the Real World: Essays Illustrating Fundamental Contributions to Society. Worth Publishers. View source
- Soderstrom, N. C., & Bjork, R. A. (2015). Learning versus performance: An integrative review. Perspectives on Psychological Science, 10(2), 176–199. View source
- Kornell, N., & Bjork, R. A. (2008). Learning concepts and categories: Is spacing the “enemy of induction”? Psychological Science, 19(6), 585–592. View source
- Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). Distributed practice in verbal recall tasks: A review and quantitative synthesis. Psychological Bulletin, 132(3), 354–380. View source
- Chen, O., Castro-Alonso, J. C., Paas, F., & Sweller, J. (2018). Undesirable difficulty effects in the learning of high-element interactivity materials. Frontiers in Psychology, 9, 1483. View source