Two alternative schemes have been proposed for coding the local direction of stimulus motion in the visual image. The "sequence discrimination" scheme (e.g. Barlow H.B. and Levick W. R., J. Physiol., Lond. 178, 477-504, 1965) uses sequential change in stimulus position over time to infer movement direction; the "spatiotemporal derivative" scheme (Marr D.M. and Ullman S., Proc. R. Soc. Lond. B211, 151-180, 1981) uses change in stimulus luminance over space and time at just one position to infer movement direction. To test these models, subjects were shown stimuli which contained combinations of stationary vertical edges and changing luminances over time. They consistently reported either leftward or rightward motion, even though no sequential change in edge location took place. Perceived directions agreed with the predictions of the spatiotemporal derivative scheme. Alternative explanations for the results based on changes in apparent edge location could not account for the data. Previous reports of apparent motion during changes in stimulus luminance are also consistent with the scheme.