Some visual stimuli produce a strong percept of motion, even though they fail to excite motion detectors based on Fourier energy or cross-correlation. Models which suffice to explain the motion percept in these non-Fourier motion (NFM) stimuli include linear spatiotemporal filtering, followed by rectification, followed by standard motion analysis (Chubb & Sperling 1988). We used the human "motion-onset" evoked potential, which has been assigned to area 17 on the basis of work in the macaque (van Dijk et al. 1986; van Dijk & Spekreijse 1989), to investigate the neural substrate of the processing stages postulated in the above models. Motion-onset VEPs elicited by FM and NFM matched for spatial and temporal characteristics were indistinguishable in temporal characteristics and scalp topography at a transverse chain of electrodes. Addition of textural cues (granularity and higher-order form) did not influence the response dynamics or scalp topography of NFM responses. However, comparison of responses to NFM stimuli and related stimuli without coherent motion but similar spatial and temporal properties showed that the motion-onset responses were distinct from responses to the onset of fixed flicker-defined contours not undergoing coherent motion. We discuss the implications of these results for computational models of motion analysis.