What do learners do when they control whether to engage in massed or spaced practice? According to theories by Son (2004) and by Metcalfe and Kornell (2005), the tendency for learners to choose spaced practice over massed practice should decline as item difficulty becomes greater. Support originally was obtained when pairs containing unfamiliar words were presented briefly for study, but subsequent research has suggested that, under these conditions, learners had difficulty initially encoding the members of the to-be-learned pairs. In Experiments 1 and 2, we failed to support the previously mentioned prediction in conditions in which the difficulty of learning was not correlated with the difficulty of initially encoding the pair members. Learners' relative preference for spaced practice increased, rather than decreased, with greater item difficulty, consistent with either a discrepancy-reduction-like account or an agenda-based-regulation account. In Experiment 3, we independently varied item difficulty and the point value that items were worth on the final test. Learners' relative preference for spaced practice was greater for high- than for low-value items but was unaffected by item difficulty. These results are more consistent with an agenda-based-regulation account than with a discrepancy-reduction account. More generally, learners' choices appear to be strategic and to reflect theory-based decisions, suggesting some level of appreciation for the relative benefits of massed versus spaced practice. (Contains 1 table, 3 figures and 2 footnotes.)