Illusory contours (ICs) are thought to be a result of processes involved in the perceptual recovery of occluded surfaces. Here, we investigate the relationship between real and illusory contour perception using a shape discrimination task and backward masking paradigm. ICs can mask other ICs when times between mask onset and stimulus onset, or SOAs, are very long ( approximately 300 ms), but real contours (RCs) are not similarly effective. Masking is absent for RC masks at perceptually salient contrasts, as well as for those with contrast lowered to match the perceived brightness of the illusory surface. We also find that RCs are not masked at long SOAs, either by ICs or by other RCs. Finally, the masking seen between ICs can occur for different sizes of target and mask. The cross-size masking would not be expected if the masking were at a level sensitive to retinal contour location. The late masking therefore may be related to a higher level of processing of shape categories and surfaces, the level at which shapes defined by ICs and RCs are differentially represented.