Practice, in the form of movement repetition, is central to motor learning. However, we do not know how consistently a person must move in order to benefit from repeated practice. We aimed to answer this question through a combination of computational modeling and behavioral experiments. We found that the total magnitude of use-dependent learning was dependent on practice consistency, but this dependence faded quickly, leaving a small but sustained aftereffect that is resistant to completely washing out. A simple model that learned only from recent movement history could explain our results. These findings should have broad implications for the study of locomotor learning as repetition is central to both basic locomotor learning and gait rehabilitation studies.