When performing successive recall tests without restudy, subjects' recalls exhibit marked variability across tests. When examining the cognitive mechanisms underlying this variability, research has focused primarily on hypermnesia, the finding that recall performance increases across tests (Erdelyi & Becker, 1974). Studies of hypermnesia usually distinguish between mechanisms supporting recall of items gained across tests versus recall of items maintained across tests, or examine associative processes of all recalled items. The present study of immediate free recall and final free recall examines associative processes supporting item gains and maintained items separately, leveraging recent clustering analyses of temporal and semantic contiguity (Polyn, Norman, & Kahana, 2009). For items studied in the same list, temporal clustering and semantic clustering for maintained items and even for gained items in final free recall than immediate free recall. Transitions in final free recall between items from different lists also reflected temporal and semantic associations. However, in final free recall a maintained item was more likely to be recalled after another item from the same list than an item gain. Further, subjects exhibiting greater list-level temporal clustering maintained more items across tests, but list-level temporal clustering did not correlate positively with item gains. The results highlight the importance of episodic and semantic associations to changes in recall across tests, and have implications for current theories of hypermnesia.