Large errors in the perception of vertically are generated by luminance borders (integrated across space) not by subjective borders.

The rod-and-frame illusion shows large errors in the judgment of visual vertical in the dark if the frame is large and there are no other visible cues (Witkin and Asch, 1948 Journal of Experimental Psychology 38 762-782). Three experiments were performed to investigate other characteristics of the frame critical for generating these large errors. In the first experiment, the illusion produced by an 11 degrees tilted frame made by luminance borders (standard condition) was considerably larger than that produced by a subjective-contour frame. In the second experiment, with a 33 degrees frame tilt, the illusion was in the direction of frame tilt with a luminance-border frame but in the opposite direction in the subjective-contour condition. In the third experiment, to contrast the role of local and global orientation, the sides of the frame were made of short separate luminous segments. The segments could be oriented in the same direction as the frame sides, in the opposite direction, or could be vertical. The orientation of the global frame dominated the illusion while local orientation produced much smaller effects. Overall, to generate a large rod-and-frame illusion in the dark, the tilted frame must have luminance, not subjective, contours. Luminance borders do not need to be continuous: a frame made of sparse segments is also effective. The mechanism responsible for the large orientation illusion is driven by integrators of orientation across large areas, not by figural operators extracting shape orientation in the absence of oriented contours.