The angular vestibulo-ocular reflex (aVOR) stabilizes retinal images by counter-rotating the eyes during head rotations. Perfect compensatory movements would thus rotate the eyes exactly opposite to the head, that is, eyes vs. head would exhibit a unity gain. However, in many species, but also in elderly humans or patients with a history of vestibular damage, the aVOR is far from compensatory with gains that are in part considerably lower than unity. The reason for this apparent suboptimality is unknown. Here, we propose that low VOR gain values reflect an optimal adaptation to sensory and motor signal variability. According to this hypothesis, gaze stabilization mechanisms that aim at minimizing the overall retinal image slip must consider the effects of (1) sensory and motor noise and (2) dynamic constraints of peripheral and central nervous processing. We demonstrate that a computational model for optimizing retinal image slip in the presence of such constraints of signal processing in fact predicts gain values smaller than unity. We further show specifically for tadpoles of the clawed toad, Xenopus laevis with particularly low gain values that previously reported VOR gains quantitatively correspond to the observed variability of eye movements and thus constitute an optimal adaptation mechanism. We thus hypothesize that lower VOR gain values in elderly human subjects or recovered patients with a history of vestibular damage may be the sign of an optimization given higher noise levels rather than a direct consequence of the damage, such as an inability of executing fast compensatory eye movements.