We examine the role of visual feedback in the programming and execution of reaching movement in patients with Parkinson's disease without cognitive impairment and patients with Alzheimer's disease without extrapyramidal signs. Controls were normally aging subjects. All subjects moved a cursor to targets on a digitizing tablet without seeing their limb. Starting and target positions were always visible on a screen while, during movement, cursor position was either visible or blanked. They were instructed to make uncorrected movements, as fast and as accurate as possible without minimizing reaction time. In absence of visual feedback, movement accuracy in patients with AD was severely impaired. Hand paths of parkinsonian patients were as accurate as normal subjects' with similar temporal velocity profiles and movement speed. With cursor feedback, accuracy was the same in the three groups, although movement speed and transport phase in patients with Alzheimer's disease were significantly reduced compared to the other groups. Also, movements of parkinsonian patients showed shorter transport phase and lower mean velocity than controls'. The different characteristics of the motor performance suggests that in the two diseases visual information is used differently for both motor programming and execution: patients with Alzheimer's disease, while scarcely using feed forward commands, relied on continuous on-line external cues. The correlation of motor performance with cognitive impairment argues against the hypothesis of basal ganglia involvement in AD. The motor abnormalities we found may represent early subclinical manifestation of apraxic disturbance. Parkinsonian patients showed higher reliance on feedback commands only with cursor feedback: this could be explained by their difficulty in engaging effectively automatic routines when distractors are present.