Saccadic eye movements are both fast and accurate. Despite the changes in the oculomotor system brought about by growth, injury and aging, saccades continue to remain accurate until we are well into our 70s. Lesion studies in both humans and monkeys show that the cortex of the cerebellum is crucial in this motor learning. One attractive model suggests that learning takes place at synapses on cerebellar Purkinje (P-) cells, which discharge both the infrequent complex spike (CS) and the high frequency simple spike (SS). In this scenario, a persistent motor error increases the probability of CS occurrence, which in turn causes changes in SS activity such that its downstream effects are appropriate to reduce the error. When we deceive monkeys into thinking their targeting saccades are in error by jumping a target forward or backward so targeting saccades either under- or overshoot respectively, the oculomotor system eliminates the dysmetria gradually over several hundred repeated deceptions. Equipped with this behavioral paradigm, we have already shown that CS activity, which starts about 80 ms after a dysmetric saccade, reports both the direction and magnitude by which a saccade is in error. We ask here whether this CS activity actually drives this saccade adaptation by inserting an artificial error signal into the CS pathway when no visual error in fact exists. Anatomical studies have shown that the CS saccade-related pathway originates in the superior colliculus (SC). Therefore, we deliver brief sub-threshold stimulus trains ˜ 80 ms after a targeting saccade lands to different sites in the topographically organized SC to cause putative errors with a variety of vectors. As the saccade is underway, we turn out its target so the only CS error signal is created by sub-thresh SC stimulation. This SC stimulation causes saccade adaptations with characteristics that are remarkably similar to those of behavioral adaptations produced using vector-matched visual errors. In particular, stimulus-induced adaptation is gradual, can produce both amplitude increases and decreases depending on the direction of the targeting saccade, and shows learning retention and gradual recovery. Moreover, the stimulus-adapted saccade can made to undergo changes in direction and depends on the timing of the stimulus-induced error. We conclude that CS activity actually drives saccade adaptation and likely also serves as the error signal in the learning of all precision behaviors as well.