Reaching towards a visual object in the absence of visual referents relies on a chain of information, from the sensory signals encoding the object's image on the retina, to the motor signals driving the hand. One link in this chain is an extraretinal eye position signal (EEPS), which specifies the position of the eye in the head. EEPS must be updated in precise coordination with the eye's rapidly changing position, or perisaccadic visual targets will be mislocalized. There have been conflicting reports about the existence and nature of mislocalizations associated with saccades. We measured perisaccadic visual localization by presenting brief (250 microseconds), bright (6000 cd/m2), binocular, gaze-point (foveal) probe flashes in an otherwise dark field to normal human subjects instructed to point to them with an unseen hand. Saccade and fixation targets were auditory, making intravisual comparison impossible. Saccades, elicited randomly to left and right of straight ahead, had a mean magnitude of 8.9 deg. Control trials, employing only non-perisaccadic probes and providing feedback of pointing errors, were randomly interspersed, to monitor and control drift of hand-eye coordination. On average, localization began to shift for probes presented 2 msec after the eye began to move, reaching a stable post-saccadic value with time constant tau = 71 msec. A second experiment was similar, except that viewing was monocular, and probes were presented randomly, at gaze (on fovea), 6 deg left of gaze (right of fovea) and 6 deg right of gaze (left of fovea). The main analysis treated saccades larger than 8 deg: their mean magnitude was 12.9 deg. Flashes left of gaze were relocalized faster (tau = 65 msec) than flashes right of gaze (tau = 129 msec) around the time of leftward saccades. In contrast, flashes right of gaze were relocalized faster (tau = 62 msec) than flashes left of gaze (tau = 90 msec) around the time of rightward saccades. Time constant was independent of saccade size. Updating began for probes presented within 4 msec of the beginning of saccades, and was not a function of saccade or flash direction. Thus, there were no systematic mislocalizations of probes presented before eye movement, and large mislocalizations of probes presented during and after. Mislocalizations were, on average, always in the direction opposite the saccade, and were maximal (about half the magnitude of the completed saccade) near the end of the saccade. Stable post-saccadic localization was not achieved until about 100-300 msec after completion of a saccade; EEPS was updated slowly, compared to eye position itself. The visual field was not remapped uniformly: the side that would normally contain the target of a visually evoked saccade (and usually the target of a corrective saccade), was updated with a shorter time constant.