1. Single-unit activity was recorded from the superior colliculus (SC) of monkeys trained to look to visual targets presented on an oscilloscope screen. On one task, target localization required that information concerning the retinal position of the target be combined with information concerning current or future eye position. This task also permitted a dissociation between the site of retinal stimulation and the metrics of the saccade triggered by the stimulation. 2. Vigorous visual responses of superficial SC neurons may occur that do not result in the activation of underlying saccade-related cells. The activity of these neurons signals the occurrence of a visual stimulus, whether or not the stimulus is selected for foveal viewing. 3. Saccade-related (SR) discharges of most intermediate and deep-layer SC neurons precede saccades with particular vectors, regardless of the region of retinal activation initiating the saccade. The discharge of these neurons is tightly coupled to saccade onset, even if changes in eye position have occurred since target appearance. Thus, the discharge of these SR neurons must occur after retinal error and eye-position signals have been combined to compute the necessary saccade vector. For most SR neurons, direct retinal activation of overlying visual neurons had no effect on either the vigor or probability of a SR discharge. The discharge of overlying visual cells is neither necessary nor sufficient to activate most SR cells. 4. The discharge of some SR cells is dependent on prior activation of overlying visual cells. Of 53 SR cells, only 3 were completely dependent on visual stimulation, while another 8 discharged less vigorously if corresponding visual activation failed to occur. 5. About one-quarter of the SR cells showed long-lead preburst activity. This activation was characterized by a low level of firing, which began after the saccade signal and continued until a saccade-linked burst occurred. 6. Cells were isolated that were visually responsive yet discharged prior to saccades in the absence of appropriate retinal stimulation. No component of the discharge of these quasi-visual (QV) cells appeared to be motor in the usual sense. The activity of these neurons appears to reflect eye-position error (the difference between actual and desired eye position) and to hold this information in spatial register until a saccade occurs or is cancelled. 7. It is concluded that the presumed linkage, implied in earlier versions of the foveation hypothesis, between the superficial layers (receiving direct retinal inputs) and the deeper layers of the SC is not necessary for the activation of SR neurons. Results suggest that the SC must generate or receive a signal that combines retinal error and eye-position information. These findings are discussed in terms of current models of the saccadic-control system.