1 Introduction Perception requires the brain to produce a coherent and unique percept on the basis of ambiguous sensory information. Because the input often underspecifies the nature of the stimulus, our perceptual systems sometimes give rise to illusions and other sorts of errors öfor example, when sitting in a stationary train while watching the neighboring train pull out of the station. This phenomenon is called visually induced self-motion, or vection (Mach 1875; Fischer and Kornmu« ller 1930). This sort of linear vection (Young et al 1973; Berthoz et al 1975; Pavard and Berthoz 1977) occurs because whole-scene motion signals induce self-motion perception, and the sensory inputs arising from steady motion of the self through a stationary environment are in fact identical to those produced by a moving scene past a stationary observer. In addition, rotating patterns produce, in the simplest case, circular vection, which can be elicited when the surrounding pattern rotates around the mid-body axis of an upright observer. In this case, observers feel as if they themselves are rotating in the opposite direction to the actual rotation. Moreover, rotations of the scene around the line of sight produce roll vection. Dichgans et al (1972) used a display of random texture, which contained no orientation cues to the visual vertical to produce such roll vection; this stimulation induces the visual sensation that the observer is continu- ously rotating, which conflicts with extraretinal gravity-receptive information signaling no change in posture. This conflicting sensory information results in an illusory shift of the internal representation of the direction of gravity, which leads observers to feel that their bodies are tilted öand also leads observers to perceive a vertical line as tilted in the same direction. Studies in weightlessness have shown even stronger effects of roll-vection stimulation than occur when gravity is present (Young et al 1986a, 1986b). Studies of vection generally emphasize the role of sensory stimulus properties, and focus almost exclusively on bottom ^ up processing. For example, much has been discovered about the role of velocity and spatial frequency (Lestienne et al 1977), area and size of stimulation (Brandt et al 1973), relative motion and distance between foreground and background (Howard and Heckmann 1989; Howard and Howard 1994; Nakamura and Shimojo 1999), and the type of visual motion pattern used to elicit vection (Andersen and Braunstein 1985). In addition, the neural mechanisms underlying these Mental imagery of visual motion modifies the perception of roll-vection stimulation Perception, 2001, volume 30, pages 945 ^ 957 Fred W Mast, Alain Berthozô, Stephen M Kosslyn Department of Psychology, Harvard University, William James Hall, 33 Kirkland Street, Cambridge, MA 02138, USA; e-mail: fmast@wjh.harvard.edu; ô Laboratoire de la Physiologie de la Perception et de l'Action, CNRS et Colle© ge de France, 11 place Marcelin Berthelot, F 75005 Paris, France Received 8 June 2000, in revised form 2 May 2001 Abstract. When viewing a wide-angle visual display, which rotates in the frontoparallel plane around the line of sight, observers experience an illusory shift of the direction of gravity; this shift leads to an apparent tilt of the body and displaces allocentric space coordinates. In this study, subjects adjusted an indicator to the apparent horizontal while viewing a rotating display. To determine whether top ^ down processes could affect the illusion, the subjects were asked to visualize a rotating configuration of dots onto a blank central portion of the moving visual field. Visualizing dots and actually viewing the dots deflected the spatial judgment in very similar ways. These results demonstrate that top ^ down processing can affect allocentric space coordinates. DOI:10.1068/p3088