Previous studies have shown that a saccade is coded in a specific reference frame according to its goal: to aim for a new object or to explore an object which has already been fixated. In a two saccade sequence, the second saccade aiming for a new object is programmed in a retinocentric reference frame in which the spatial location of the second object is stored in spatial memory before the first saccade and updated after its execution. The second saccade exploring the same object is coded in an oculocentric reference frame in which object size is directly transformed into a fixed motor vector, encoded in motor memory before the first saccade and simply applied after its execution. The integration of parafoveal visual information appears to be crucial in the selection of the appropriate reference frame. The two experiments presented here investigate how and when the saccadic system integrates visual information to plan a sequence of saccades. In separate blocks, subjects were asked to execute a sequence of prosaccades directed toward a single object or two short objects, or to execute a sequence of antisaccades in the opposite direction of the stimuli. The latency of the initial saccade was modulated by using the Gap-200, Gap-0 and Overlap-600 ms paradigms. The results show that the time available for segmenting the visual stimulation into discrete objects and application of a specific reference frame according to this segmentation is critical for saccadic planning.