Prosaccades and antisaccades were investigated in three groups of patients with parkinsonian syndromes, Parkinson's disease, corticobasal degeneration (CBD) and progressive supranuclear palsy (PSP), and in a control group. Saccade tasks were performed in single-task blocks (i.e. either blocks of prosaccades or blocks of antisaccades) and in mixed-task blocks (i.e. in blocks of randomly interleaved pro- and antisaccades). Saccade latencies and directional errors (misdirected saccades) were analysed in each subject, and we concentrated more specifically on the comparison of error rates in single tasks and in repeated trials of mixed tasks (i.e. mixing costs). The performance of each group in single tasks was largely consistent with previous studies, with normal antisaccade error rates in Parkinson's disease and CBD patients and increased antisaccade error rates in PSP patients. In contrast, a double dissociation was observed in mixed tasks. Parkinson's disease and CBD patients showed a marked increase in prosaccade and antisaccade error rates in repeated trials of mixed tasks, illustrated by increased mixing costs, whereas PSP patients showed similar error rates in single and repeated trials of mixed tasks, i.e. normal mixing costs. These results demonstrate that: (i) antisaccade performances may be differentially affected in mixed tasks and single tasks; (ii) the region of the dorsolateral prefrontal cortex which is crucial for reflexive saccade inhibition does not seem to be involved in the additional processes required in mixed-task conditions; (iii) the study of interleaved pro- and antisaccades may increase the accuracy of the differential diagnosis between these parkinsonian syndromes.