Depth finding and surface detection will be important factors in robot vision, though it is not clear whether robot eyes will be monocular and use projected grids or reflected laser beams, or whether they will simulate nature using calculations from a stereo pair. This report probes deeply into the binocular approach for images with textured surfaces and with discontinuities of surface, depth, and orientation. The main technical innovation is the simultaneous use of pointwise matching in the stereo pair and estimation of surface patches to reduce the amount of computation. The paper gives a large amount of technical description of methods, much of which refers to other basic works in the field. These methods are, by necessity, becoming hard to describe in full detail. The authors illustrate results for 12 simple scenes. The bibliography contains 30 references.