When a person walks through a rigid environment while holding eyes and head fixed, the pattern of retinal motion flows radially away from a point, the focus of expansion (Fig. 1a). Under such conditions of translation, heading corresponds to the focus of expansion and people identify it readily. But when making an eye/head movement to track an object off to the side, retinal motion is no longer radial (Fig. 1b). Heading perception in such situations has been modelled in two ways. Extra-retinal models monitor the velocity of rotational movements through proprioceptive or efference information from the extraocular and neck muscles and use that information to discount rotation effects. Retinal-image models determine (and eliminate) rotational components from the retinal image alone. These models have been tested by measuring heading perception under two conditions. First, observers judged heading while tracking a point on a simulated ground plane. Second, they fixated a stationary point and the flow field simulated the effects of a tracking eye movement. Extra-retinal models predict poorer performance in the simulated condition because the eyes do not move. Retinal-image models predict no difference in performance because the two conditions produce identical patterns of retinal motion. Warren and Hannon observed similar performance and concluded that people do not require extra-retinal information to judge heading with eye/head movements present, but they used extremely slow tracking eye movements of 0.2-1.2 deg s-1; a moving observer frequently tracks objects at much higher rates (L. Stark, personal communication). Here we examine heading judgements at higher, more typical eye movement velocities and find that people require extra-retinal information about eye position to perceive heading accurately under many viewing conditions.