Two hypotheses proposed as alternatives by Rock--frame of reference and hierarchical organisation of perception--were tested in a series of experiments with the use of the rod-and-frame illusion. This illusion produces errors in the apparent vertical due to the presence of a tilted frame surrounding the test rod. The apparent vertical is shifted in the direction of the frame tilt. When an upright square was added inside the tilted frame, rod-setting errors varied according to the visual characteristics of the display. In the case of a large display presented in the dark (experiment 1), there continued to be large errors in the direction of the outer-square tilt. This finding supports the frame-of-reference hypothesis, which proposes that the orientation of all objects in the visual field is dominated by the most peripheral reference. In the case of a small display presented in a lit environment (experiments 2 and 3), the direction of errors was the opposite. This latter finding was taken to indicate that the rod was set with reference to the perceived tilt of the inner upright square. Thus, according to a hierarchical-organisation hypothesis, the orientation of an object in the visual field is influenced by objects in the immediate surroundings not by outermost reference. Overall, the results confirm the presence of two qualitatively different classes of orientational phenomena: one is concerned with the definition of egocentric coordinates and one with an object-centred visual representation.