In a series of experiments in which stereo and monocular 3D interpretations were made mutually inconsistent, stereopsis was found to integrate with monocular depth only when disparities varied nonlinearly spatially. The apparent spatial disposition, including local surface orientation and depth ordering, was dominated by the monocular interpretation, even when stereo infor­mation provided strongly contradictory information. We conclude that stereo and monocular vision is integrated primarily on the basis of topographic features, not scalar depth values. More­over, in interpreting depth from stereograms, we found a stereoscopic analogue to the familiar brightness contrast effect. The depth interpretation of stereo disparity information is roughly analogous to edge detection, with insensitivity to constant disparity gradients analogous to our insensitivity to linear (ramp-like) intensity distributions. This rules out several current models of spatial integration, and suggests new computational strategies.