Previous studies underline the role of dopamine in cognitive reinforcement learning. This has been demonstrated by a striatal involvement in feedback-based probabilistic classification learning. In order to determine to which extent the dopaminergic loss of Parkinson's disease and aging affects the feedback aspect in classification learning, we applied two versions of the same visual classification task. One version had to be learnt by trial-by-trial feedback, the other by observing the correct assignment of stimulus and category. Performance was evaluated in test blocks that were identical under the feedback and the observational conditions. There were 31 patients with Parkinson's disease (PD), 30 older controls and 20 younger controls tested. The results show that younger healthy participants perform better than older participants in the classification task and this difference significantly interacts with the learning condition: both groups show nearly the same level of performance under the observational condition but younger participants show a better performance than older ones under the feedback condition. In contrast, PD patients and older controls did not differ in their performance in the classification task; both groups performed better under the observational than under the feedback condition. These results demonstrate that healthy aging affects feedback-based learning but does not affect learning by observation. The fact that PD patients showed no additional deficit in feedback-based learning is an indication that the loss of dopamine does not play the key role under the feedback condition of our classification task. This finding questions the general role of the striatum in feedback-based learning and demonstrates that healthy aging particularly affects feedback-based learning.