NEURAL events underlying perception of coherent motion are generally believed to be hierarchical(1,2) : information about local motion is registered by spatio-temporal coincidence detectors(3-5) whose outputs are cooperatively integrated at a subsequent stage(6,7). There is disagreement, however, concerning the spatial scale of the neural filters underlying these operations. According to one class of models, motion registration is initially accomplished in parallel at multiple spatial scales(3-5), with filters tuned to lower spatial frequencies responsive to larger motion displacements than filters tuned to higher frequencies. According to another scheme, motion analysis involves a single, broadly tuned spatial filter, with optimal displacement dependent on spacing of local elements(8). Here we use a masking procedure to measure the extent to which dynamic noise depicted at one spatial scale interferes with detection of coherent motion conveyed by image features at another spatial scale. Our results indicate that a single filter, broadly tuned for spatial frequency, is mediating detection of coherent motion. This finding dovetails with known physiological properties of neurons at an intermediate stage of motion processing.