Many insects show by their behaviour that they detect visually the existence of separate objects. The experimental material to analyse how they perceive objects is provided by an insect that walks to the end of a stick; then, because it has no alternative, it reaches with a foreleg towards a neighbouring object that it perceives to be within range. Some insects make horizontal peering movements as an aid to vision. The peering motion is exactly appropriate for generating an apparent velocity of nearby objects relative to the background. These experiments, when put together with the known properties of optic lobe neurons, suggest that a mechanism based on velocity parallax projected to the horizontal plane accounts for much insect visual behavour. Velocity parallax is defined as the discrepancy seen at the edge of an object against a distant background when the eye moves laterally. On this theory, perception of an object is inseparable from the local detection of velocity differences. The background may not be `perceived' at all when an object occurs in the foreground. The postulated mechanism is a two- or three-stage feedback, in which the perceived velocity (or, more accurately, the spatially correlated contrast frequency) in small-field motion-perception units is reduced by the averaged contrast frequency in larger fields, which feed back upon them. Contrast frequency is defined as the frequency of the flicker that is generated by a pattern moving across the eye. An alternative mechanism to the feedback of the velocity signal with lateral spread is adaptation to the local average background velocity, while sensitivity to a smaller local change in velocity is retained. That idea comes from recent work on the H1 neuron in the fly optic lobe, and could be the basis of a primitive form vision that, if present in medium-field neurons, is adequate for the whole of the normal visual behaviour of a freely moving insect. These speculations invite a variety of experimental tests, ranging from visual discrimination tests with bees that are shown the velocity parallax situation, to appropriate stimulation of optic lobe neurons, to simulation of a visual processing system that relies on velocity parallax cues to detect objects.