We previously reported that Parkinson's disease patients could point with their eyes closed as accurately as normal subjects to targets in three-dimensional space that were initially presented with full vision. We have now further restricted visual information in order to more closely examine the individual and combined influences of visual information, proprioceptive feedback, and spatial working memory on the accuracy of Parkinson's disease patients. All trials were performed in the dark. A robot arm presented a target illuminated by a light-emitting diode at one of five randomly selected points composing a pyramidal array. Subjects attempted to "touch" the target location with their right finger in one smooth movement in three conditions: dark, no illumination of arm or target during movement; movement was to the remembered target location after the robot arm retracted; finger, a light-emitting diode on the pointing fingertip was visible during the movement but the target was extinguished; again, movement was to the remembered target location; and target, the target light-emitting diode remained in place and visible throughout the trial but there was no vision of the arm. In the finger condition, there is no need to use visual-proprioceptive integration, since the continuously visualized fingertip position can be compared to the remembered location of the visual target. In the target condition, the subject must integrate the current visible target with arm proprioception, while in the dark condition, the subject must integrate current proprioception from the arm with the remembered visual target. Parkinson's disease patients were significantly less accurate than controls in both the dark and target conditions, but as accurate as controls in the finger condition. Parkinson's disease patients, therefore, were selectively impaired in those conditions (target and dark) which required integration of visual and proprioceptive information in order to achieve accurate movements. In contrast, the patients' normal accuracy in the finger condition indicates that they had no substantial deficits in their relevant spatial working memory. Final arm configurations were significantly different in the two subject groups in all three conditions, even in the finger condition where mean movement endpoints were not significantly different. Variability of the movement endpoints was uniformly increased in Parkinson's disease patients across all three conditions. The current study supports an important role for the basal ganglia in the integration of proprioceptive signals with concurrent or remembered visual information that is needed to guide movements. This role can explain much of the patients' dependence on visual information for accuracy in targeted movements. It also underlines what may be an essential contribution of the basal ganglia to movement, the integration of afferent information that is initially processed through multiple, discrete modality-specific pathways, but which must be combined into a unified and continuously updated spatial model for effective, accurate movement.