Previous research has revealed that people choose to aim toward an "optimal" endpoint when faced with a movement task with externally imposed payoffs. This optimal endpoint is modeled based on the magnitude of the payoffs and the probability of hitting the different payoff regions (endpoint variability). Endpoint selection, however, has only been studied after people had experience with the aiming task. The present study examined initial endpoint selection and how it changed as a function of experience with performing the task. Participants completed 300 movements to a target that was overlapped by a penalty region. Mean endpoint was analyzed in intervals of 50 trials. Predictions based on the optimal model would indicate that the mean endpoint should be farther from the penalty region early in practice when endpoint variability is higher--increasing target misses but decreasing costly penalty hits. As variability decreases, however, it is predicted that the endpoint should shift closer to the optimal location. In contrast to these predictions, participants' mean movement endpoints started closer to the penalty region and shifted away with increasing practice, even as endpoint variability decreased. This pattern of endpoint selection leads to suboptimal gains early in experience but more optimal gains by the end of the trials. These findings suggest that, similar to results found in cognitive decision-making tasks, people need to receive performance-based feedback to weight their estimations of probability and payoffs in motor tasks. (Contains 1 table and 4 figures.)