Shading of 3-D objects refers to variations in luminance or color as a function of the angle of the illumination direction with respect to the surface. Shading gives rise to a compelling sense of shape, illumination conditions and surface properties. However, it is not yet certain how this is achieved. We used an illuminant direction matching procedure to examine how well human observers can estimate the direction of illumination in images of rotating 3-D geometrical shapes in different light fields. Stimuli were renderings of a dodecahedron and a sphere projected orthographically and juxtaposed on a CRT screen. Two lighting modes (collimated and hemispherical diffuse lighting), twenty-six illuminant directions (differing in azimuth and elevation) and perfectly diffuse reflectance with albedo 0.5 were used to render the faces of the two stimuli. Since the two shapes were convex and situated in empty space, neither interreflection nor cast shadow was produced. At the start of each trial, the two stimuli were presented with different directions of illumination. Observers adjusted the direction of illumination on the sphere to match the direction of illumination on the dodecahedron. Results revealed no systematic difference in matched directions between the two lighting conditions, neither in azimuth nor in elevation. This suggests that in the absence of cast shadow, observers may use the location of the brightest point as a cue to the direction of illumination. Although the appearance of image shading can be appreciably changed under different lighting conditions, for the case of Lambertian reflectance, the peak in the luminance distribution remains fixed.