We previously reported a computer model to reproduce the perception of the Kanizsa figure and a wide range of variation figures, based on depth cue detection and surface reconstruction algorithms (Kogo et al., 2002, Lecture Notes in Computer Science, 2525, 311-321). In our model, the local depth cues are first detected and then globally integrated in a process of surface construction to determine the 3D structures. The model predicts that this 3D information plays a key role in the different perceptions of the variation figures. To support this view, we first revisited the psychophysical experiments (Halpern, In Petry & Meyer, 1987, The perception of illusory contours, pp171-175; Guttman & Kellman, 2004, Vision Research, 44, 1799-1815), that have been reported to measure the brightness perception in the central area (brightness-nulling experiment), and the perceived positions of the subjective contours (contour positioning experiment). As predicted, the Kanizsa figure showed the perceived brightness being different from the surrounding area, and the positions of the subjective contours were perceived more inwardly than in the figures with no illusion. The variation figures that do not evoke the illusion did not result in these properties. Importantly, the figures that consist of an equal number of white and black objects on a neutral grey background also showed the contour positions inwardly as well, while the brightness was not different from the surrounding area. Next, using a stereoscopic setup, subjects were asked to change the disparity of the central area until it was perceived to have the same depth as the surrounding objects (depth-nulling experiment). The brightness-nulling and contour positioning experiments were repeated after the subject found the nulling position in depth. We compared the measurements of the perceived brightness and contour positions before and after the depth-nulling and evaluated the results against our model predictions.