Short answer: further apart than feels comfortable, and the right distance depends on when you are tested.
What is settled
That spacing beats massing is among the most reliably replicated findings in the psychology of learning. It has held across hundreds of experiments, across age groups, across material types, and across gaps ranging from minutes to months. Dunlosky and colleagues (2013) gave distributed practice one of only two high utility ratings among the ten techniques they assessed.
If you take one thing from the research on studying, take this one. The same hours, rearranged, produce more learning. Nothing else on offer is that cheap.
The study that put numbers on it
Cepeda, Vul, Rohrer, Wixted and Pashler (2008) taught a set of facts to more than 1,350 people, brought them back for a review after a gap of anywhere up to three and a half months, then tested them at a further delay of up to a year. That design is unusual: most spacing research uses intervals of minutes or days, which tells you little about preparing for something in June.
What they found is that as the gap between study and review widens, performance first improves and then declines. There is a peak. And the location of that peak moves with how long you need to remember the material.
Measured as a proportion of the delay until the test, the optimal gap fell from roughly twenty to forty per cent when the test was a week away, to roughly five to ten per cent when it was a year away. For a test seventy days out, the best-performing gap in their data was around three weeks.
How precise is any of this
Less precise than a percentage makes it sound, and that is good news rather than bad.
The peak is broad. Performance near the optimum changes slowly, so landing somewhat off it costs very little. The expensive mistake is not choosing seventeen per cent when twenty-two would have been better. The expensive mistake is a gap near zero, which is what studying the same thing on consecutive evenings amounts to, and which the same data show performing worst of all.
So treat any figure here as a useful starting point rather than a prescription. The difference between a good gap and a perfect one is small. The difference between spacing and not spacing is not.
What to actually do
- Exam in a week: review tomorrow or the day after, then again a few days later.
- Exam in a month: first review after four or five days, then roughly double each gap.
- Exam in six months: first review after two or three weeks, then widen from there.
- Exam in two days: spacing cannot help you now. Spend the time on retrieval practice and sleep the full night.
- When in doubt, go longer. Erring long costs less than erring short, because a gap that is too short is barely different from not spacing at all.
The spacing calculator on this site does this arithmetic for you, including the widening, and will put the dates in your calendar or send them to you.
The part people skip
A review is a closed-book attempt to recall the material. It is not a reread of it.
Spacing rereads produces a fraction of the benefit, because you are spacing the wrong activity. The two high-utility techniques in Dunlosky’s review are distributed practice and practice testing, and they are at their best combined: retrieval attempts, spread out. Take the schedule and use it to reread on a timetable and you have adopted the inconvenient half of the method while leaving out the half that does the work.
If a scheduled review feels easy, that is information. It means the gap was too short, or you are recognising rather than recalling. Both are fixable, and both are invisible if you never close the book.
In the book
Chapter 5 covers the evidence in detail. Chapter 4 covers what a review should consist of. Chapter 27 covers fitting this into a semester rather than a single subject. Appendix B lists the studies in full.