When an animal's surroundings move, the animal normally follows that movement by turning its eyes (that is, by an optomotor reaction). As a result, the retinal image is partly stabilized. The efficacy of this stabilization necessarily depends on the gain of the optomotor control circuit. So far no biological detectors of retinal image movements have been discovered in either vertebrates or invertebrates that is, elements capable of generating a signal proportional to the movement velocity, which could serve as sensors in this control system (Borst & Egelhaaf 1989). The reason is that many other parameters, such as the light intensity and the `texture' of the pattern, also affect the neuronal output. If movement detection is texture dependent, for instance, the gain and hence the quality of stabilization must also be texture dependent. But in humans, at least, with large-field stimulation the quality of retinal image stabilization has been found to be largely independent of texture (de Graaf et al. 1990). Here I describe a control system with gain control that permits automatic compensation, under closed-loop conditions, of the dependence of movement detection on parameters such as texture, brightness and so on. Comparison with data from experiments on arthropods shows that, in these animals at least, a control circuit with nonlinear properties like those suggested here has in fact been realized.