Seventy-two pigeons were given single-stimulus training for variable-interval reinforcement and were then tested for generalization (preference) with a white line on a dark surround at angles of 90 (vertical), 60, and 30 deg. The training stimuli were (in different groups) a white light, a very dim white light, a green light, and a white dot on a dark surround. Any preexperimental preference for a given line angle should have been reflected in all groups; however, no such consistent preference was found. All training stimuli were logically orthogonal to the line angle dimension, yet training with the white light or the dim light resulted in a preference for vertical, training with the dot resulted in a preference for 30 deg, and training with the green resulted in no preference at all. Thus, orthogonality is an empirical rather than a logical matter, and it must be demonstrated in the context in which the concept is to be employed.