The motion of a visual target in the periphery activates visuomotor channels and begets an interceptive saccade. In response to a transient motion, the landing position depends upon the target speed. It falls short but linearly increases with the landing time if the target accelerates. Moreover, later saccades land beyond the location where the target disappeared (Quinet & Goffart 2015). Here, we artificially delayed the landing time and tested, in three rhesus monkeys, the persistence of visual signals and the time course of their influence upon the saccade system. After fixating a central target, a target briefly moved (150-200 ms) leftward or rightward, in the upper or lower visual field. The monkeys were rewarded after each saccade toward the correct quadrant. For each trial, the target path was pseudo-randomly selected and its instantaneous speed increased from 0 to 40°/s. In 30-50 % of the trials, a brief microstimulation (30-50 ms) was applied in the deep superior colliculus, at target disappearance. As expected, unperturbed saccades fell short of the accelerating target and the latest ones landed beyond the location where the target disappeared. In some extremely rare trials, the landing time was late and the scatter of landing positions exhibited a “stagnation”, which was confirmed by the perturbed trials. After the stimulation-evoked saccade and a subsequent pause, the majority of correction saccades were indeed made toward a relatively fixed location. Some of them could be generated up to 500 ms after the target offset. For some stimulation sites, a decay happened with later landing times when the perturbation was ipsilateral to the target motion. Beyond revealing a persistent border for extrapolation, our study shows that perturbing a simple visuomotor reaction recruits working memory with no prior temporal training.