SACCADES of patients with mild Parkinson's disease (PD) are said to be abnormal in the absence of a concurrently visible target or when they are part of a rapid sequence of eye movements. We tested this hypothesis using a predictive saccade paradigm in which target visibility is withdrawn for a period. Three rates of target alternation were used (0.25 Hz, 0.5 Hz and 1.0 Hz). Withdrawal of target visibility brought out the extremes of primary saccade gain for both the controls and the patients with PD, most undershoot being displayed at the lowest frequency, whereas the gain was greatest at the highest frequency, actually overshooting the target location. These results demonstrate that the spatial error of parkinsonian saccades does not invariably take the form of hypometria when part of a rapid sequence of eye movements.