In daily visual scenes, familiar objects appear in various orientations, such as front, side, and oblique. How accurately can human vision determine these depth orientations? Since the extraction of the front-back axes of objects seems to involve the determination of object orientations, we predicted that the determination of object orientations would be relatively inaccurate when the front-back axes are foreshortened and difficult to extract (i.e., when the front-back axes are roughly aligned with the viewpoint). To examine this hypothesis, we presented pictures of familiar objects in various orientations (namely, azimuth angles) to participants and asked them to estimate the orientations of these objects. Ten orientations ranging from 9 to 171 degrees were tested (0 degrees indicates front orientation, and 180 degrees indicates back orientation). The objects were rendered into 2D stimulus images by using computer software and were presented on a vertical CRT screen that were placed at a distance of 500 mm from the participants’ viewpoint. This real-world 3D spatial relationship was simulated by computer software when the stimulus images were generated. The participants rotated a disk displayed on the horizontal LCD screen such that its front (marked by a dot) was oriented to the same azimuth angle as the stimulus objects. When the object orientations were less than 45 degrees or more than 135 degrees (the front-back axes were relatively foreshortened), the estimated orientations were biased toward 90 degrees (side orientation), i.e., participants overestimated the deviations from the front or back orientations. The determination of object orientations was biased because of the difficulty in extracting the front-back axes precisely. This result also suggests that human vision has adapted to determine whether or not the front-back axes of familiar objects are aligned with the viewpoint.