Theories of motor control often assume that the location of visual stimuli is expressed in non retinotopic frames of reference. The saccadic system is known in enough detail for us to examine the evidential basis of this assumption. The organization of the neural circuit that controls saccades is first summarized. It is shown to consist of at least two interconnected modules. The first one is the burst generator, which resides in the reticular formation, and is entrusted with the tasks of impedance matching, synergist coactivation and reciprocal inhibition between antagonists. The second is a metric computer, which resides in the superior colliculus and the cerebral cortex, and computes the size and direction of the desired movement. Alternative models of the burst generator are presented and their “verisimilitude” is assessed in the light of evidence concerning saccadic trajectories, neuronal discharge patterns, interneuronal connections, as well as the results of lesion and stimulation experiments. Several models of the “metric computer” in the superior colliculus are then examined; their performance is again evaluated in the light of psychophysical, anatomical, physiological, and clinical evidence. It is demonstrated that the location of visual stimuli need not be expressed in nonretinotopic frames of reference for either the burst generator or the metric computer to issue appropriate commands to move the eyes. Instead, using information concerning intervening movements of the eyes to update the location of visual stimuli in a retinotopic frame of reference suffices for the planning and execution of correct saccades. More generally, it is proposed that the location of sensory stimuli need not be expressed in higher order frames of reference (e.g., centered in the body or even in extrapersonal space) provided that their location in a sensorium specific map is updated on the basis of effector movements.