Saccade latency is reduced when the fixation stimulus is removed shortly before a saccade target appears (gap task) as compared to when the fixation stimulus remains present (overlap task). To test the assumption that this gap effect benefits from advanced motor preparation (M. Paré & D. P. Munoz, 1996), we manipulated target onset independently of the signal to launch a saccade (peripheral offset at the mirror location). In Experiment 1, we showed that, when the target appears at one of only two possible locations, target onset strongly improves performance (lower latency, higher accuracy) in the overlap task but not in the gap task. In Experiment 2, we found that the lack of an effect of target onset in the gap task was not due to inhibition of a reflexive response to the transient associated with the offset (go signal) in our task. In Experiment 3, we manipulated target onset and target uncertainty (two, four, or eight potential target locations) in gap and overlap tasks. As target uncertainty increased, the gap effect decreased, and the effect of target onset on saccade latency in the gap condition became greater. Overall, our results suggest, in line with the motor-preparation hypothesis, that saccade metrics in a gap task are computed before the target is actually displayed and that advanced motor preparation is enhanced when the location of the target is predictable. Analyses of anticipations and regular-latency errors corroborated this view.