Three experiments were carried out to determine (a) the effectiveness of associative rehearsal and rote repetition for long-term recall with and without expectation of a later recall test, and (b) the subjects' ability to assess their own future recall performance. The overall results showed that a test expectancy effect was not obtained when a voluntary selection in strategies was prohibited, but was found for subjects not instructed in the use of any particular strategy. It is also shown that the subjects could predict correctly that the items studied with associative rehearsal would be recalled more successfully at the final test than when they were studied with rote repetition. A superiority of associative rehearsal for prediction was not found when the subjects used only one type of rehearsal, but was found when they were not given the immediate test. The results suggest that the expectancy effect can be explained in terms of subjects' intentional shifts of strategies based on their metacognitive assessment of such strategies.