In three experiments we tried to mask the motions of human gait. We represented human walkers as a set of 11 computer-generated elements on a display monitor, moving as a nested hierarchy of motions that mimicked the motions of the head and major joints. The walker was seen in sagittal view, facing either right or left and walking as if on a treadmill. On the walker was superimposed a simultaneous mask composed of elements with the same brightness, shape, and subtense as those of the walker. We varied the mask parameters—particularly the number of elements and style of motion—to discern what masks best camouflaged the walker’s direction. In general, many kinds of masks impeded viewer performance at durations of 200 msec, but only relatively complex masks continued to impede performance to 400 msec and beyond. Four results stand out concerning the concurrent perceptual organization of target and mask. First, if the mask is easily divided into groups by its motion parameters, viewer performance with respect to the stimulus is generally impeded by increasing the number of groups in the mask. Second, the most successful masks are those composed of scrambled parts of walkers. Third, given a sufficient number of scrambled-walker elements, viewer performance does not improve above chance even at 800 msec. And fourth, this lack of improvement appears to be confined to scrambled-walker masks that share the particular gait parameters of the walker target.