Observers viewed computer-generated stereograms of randomly structured smooth surfaces and were required to judge the perceived local orientation at numerous probe points by adjusting a monocular gauge figure. The surfaces were depicted with specular or Lambertian reflectance functions, either with or without identifiable texture elements, and with varying directions of illumination. The results revealed a strong linear correlation between the judged patterns of relief and the actual depicted objects, though there were systematic differences in the magnitude of depth scaling in the different conditions. In general, the accuracy and reliability of observers' judgments for the smoothly shaded shiny surfaces was slightly lower than for the textured surfaces and slightly higher than for the smoothly shaded Lambertian surfaces. The direction of illumination had no detectable effect on the observers' judgments.