In Parkinson's disease (PD), it is possible that tonic increase of dopamine associated with levodopa medication overshadows phasic release of dopamine, which is essential for learning. Thus while the motor symptoms of PD are improved with levodopa medication, learning would be disrupted. To test this hypothesis, we investigated the effect of levodopa medication on learning on the weather prediction task (WPT), which involves probabilistic classification learning. 11 PD patients and 13 matched controls completed 200 trials of the WPT, with the patients either on or off their usual levodopa medication. Consistent with prior studies, when PD patients were assessed on medication, overall WPT performance was significantly worse than controls. However, when these patients were studied following withdrawal from medication, overall performance was equivalent to controls, and significantly better than when on medication. The significant deterioration of learning on the WPT in PD patients when on compared to off medication supports the proposal that tonic increase of dopamine with dopaminergic medication masks phasic changes in dopamine release essential for learning. These results highlight the need for careful "titration" of dopaminergic medication to produce the desired improvement of the motor symptoms without the associated detrimental effects on cognition and learning. (Contains 2 tables and 2 figures.)