The vertebrate retina contains only five major neuronal classes but these embrace a great diversity of discrete types, many of them hard to define by classical methods. Consideration of their spatial distributions (mosaics) has allowed new types, including large ganglion cells, to be resolved across a wide range of vertebrates. However, one category of large ganglion cells has seemed refractory to mosaic analysis: those that project to the accessory optic system (AOS) and serve vestibulocerebellar mechanisms of motion detection and image stabilization. Whenever AOS-projecting cells have been analyzed by nearest-neighbor methods, their distribution has appeared almost random. This is puzzling, because most aspects of visual processing require the visual scene to be sampled regularly. Here, spatial correlogram methods are applied to distributions of large ganglion cells, labeled retrogradely from the AOS in frogs, turtles, and rats, and to the AOS-projecting displaced ganglion cells of chickens. These methods reveal hidden spatial order among AOS-projecting populations, of a form that can be simulated either by superimposing a single regular mosaic on a random population or, more interestingly, by overlapping three or more regular, similar but spatially independent mosaics. The rabbit is known to have direction-selective ganglion cells (not, however, AOS projecting) that can be subdivided into functionally distinct, regular mosaics by their tracer-coupling patterns even though they are morphologically homogeneous. The present results imply that the direction-selective AOS-projecting ganglion cells of all vertebrates may, likewise, be subdivided into regular, independent mosaics.