In most investigations of visual masking, spatially adjacent contours are located in the same depth plane, a practice derived from the assumption that depth is not a critical variable in contour formation. In nine experiments, the dependence of masking on the relative depth positions of the target and mask was examined. To permit facile manipulation of depth without introducing confounding changes in proximal stimulation, the stimuli were stereoscopic contours formed from dynamic random-element stereograms generated as anaglyphs on a projection color-television display. The target was a briefly presented Landolt C with gap position randomly varied; the mask was an annulus that surrounded the target. Masking with stereoscopic contours was robust and subject to the same spatial and temporal variables influential in masking with physical contours. Depth separation of the target and mask produced significant reductions in masking and interacted with temporal order and egocentric stimulus position. These results have implications for models of visual masking and indicate that perceived depth is important in contour formation.