Dopamine has long been implicated in reward-based learning and the expression of such learned associations on performance. Robust evidence supports its effects on learning and performance, but teasing these apart has proved challenging. Here we have adapted a classic test of value-based learning, the probabilistic selection task, to disentangle effects of dopamine on value-based performance from effects on value-based learning. Valence-specific effects of dopamine on this specific task cannot be accounted for by modulation of learning, and therefore must reflect modulation of performance. We found that dopaminergic medication, consisting of levodopa and/or dopamine agonists taken at own dose, in 18 patients with mild Parkinson’s disease (Hoehn and Yahr < 2.5) potentiated reward-based approach in terms of both accuracy and reaction times, while leaving punishment-based avoidance unaffected. These data demonstrate that the effects of dopamine on probabilistic action selection are at least partly mediated by effects on the expression of learned associations rather than on learning itself, and help refine current models of dopamine’s role in reward.