Tests, as learning events, can enhance subsequent recall more than do additional study opportunities, even without feedback. Such advantages of testing tend to appear, however, only at long retention intervals and/or when criterion tests stress recall, rather than recognition, processes. We propose that the interaction of the benefits of testing versus restudying with final-test delay and format reflects not only that successful retrievals are more powerful learning events than are re-presentations but also that the distribution of memory strengths across items is shifted differentially by testing and restudying. The benefits of initial testing over restudying, in this view, should increase as the delay or format of the final test makes that test more difficult. Final-test difficulty, not the similarity of initial-test and final-test conditions, should determine the benefits of testing. In Experiments 1 and 2 we indeed found that initial cued-recall testing enhanced subsequent recall more than did restudying when the final test was a difficult (free-recall) test but not when it was an easier (cued-recall) test that matched the initial test. The results of Experiment 3 supported a new prediction of the distribution framework: namely, that the final cued-recall test that did not show a benefit of testing in Experiment 1 should show such a benefit when that test was made more difficult by introducing retroactive interference. Overall, our results suggest that the differential consequences of initial testing versus restudying reflect, in part, differences in how items distributions are shifted by testing and studying.