Most studies on the effects of ageing on saccades have examined reflexive saccades; the only commonly studied volitional task has been the antisaccade task, with contradictory results. We examined in both young and elderly normal subjects the latency of anti-, memory-guided, and predictable saccades and the timing of self-paced saccades; we also evaluated errors made on the first two tasks. We expected errors to be correlated between tasks; we also expected antisaccade latencies and errors to be inversely correlated. We also expected antisaccade and memory-guided saccade latencies to be longer in individuals with a high self-paced rate. Except for predictable saccades, mean latencies were significantly higher in the elderly. However, their performance was more variable. Errors were also significantly more frequent on anti- and memory-guided saccade tasks. Most of the hypothesised correlations were not observed. Analysis of error latencies showed that whilst most antisaccade errors were reflexive, for memory-guided saccades both express errors and errors with latencies between 0.4 and 2.5 s were observed. The latter appeared to be a premature release of what would otherwise have been a properly planned response. Age thus impaired all but the predictable saccade task; nevertheless, there were few relationships between measures across tasks. This suggests that a range of processes mediate peoples' performance on these saccade paradigms.