In a previous study we found large systematic errors (up to 40°) when subjects adjusted the orientation of a horizontal test bar until it appeared parallel to a horizontal reference bar, both bars rotating about their vertical axes. The deviations increased linearly with the separation angle but vanished when the orientation of the reference bar was either parallel or perpendicular to the median line. In order to test the assumption that external references caused these deviations to vanish, the same task was repeated in four different conditions: in the normal condition the horizontal aperture, formed by a cabin, and the facing wall of the room were frontoparallel to the subject; in the other conditions either the room, the cabin or both were oriented 30° to the right with respect to the subject. It was found that, depending on the subject, the occurrence of the vanishing deviations covaried with the orientation of the cabin or the room. Evidently, subjects are influenced by the external references provided by the walls of the room and the sides of the cabin. The results indicate that a description of visual space by a Riemannian metric of constant curvature is not valid in a visual environment containing external references.