The locust descending contralateral motion detector (DCMD) responds to movements anywhere within a wide visual field, but responds most strongly to the images of approaching objects. It has been claimed that the response peaks before the end of an approach movement, providing a signal that anticipates collision. However, we find that when the locust eye is presented with appropriate computer-generated images of approaching objects, the response builds up until after movement has stopped. Premature peaking in the response is due to failure to stimulate the eye with sufficiently small and frequent jumps in image edges. We conclude that the DCMD signals impending collision by tracking edge motion throughout object approach.