The authors examined the difference in errors made by eight subjects in setting a bar of light in an otherwise darkened room to either visually perceived vertical (VPV) or visually perceived horizontal (VPH) during maintained roll-tilted positions around the naso-occipital axis. Two viewing distances were examined, 25 and 60 cm. Subjects were tested at roll-tilt angles of 10 degrees intervals from upright to body horizontal (both left ear down (LED) and right ear down (RED)) in a randomized fashion. Settings were made only after a 1 min delay at each tilt angle to allow for decay of the semicircular canal signal. Chair rotation speed was 2 degrees/s with subjects being re-tested using 1/2 degree/s (at 25 cm) to determine the effect of rotation speed. Average errors for vertical versus horizontal were significantly different from each other (P < 0.01) at both the 25 and 60 cm viewing distances. The errors follow a complex function, with VPH showing smaller errors than VPV for large roll-tilts, while the opposite was true for medium-sized roll-tilts. This was true at both chair velocities. That is, VPV and VPH are not orthogonal to one another under the conditions examined. There are large differences between individuals but each individual showed a repeatable pattern. The average extent of non-orthogonality was found to be as high as 7 degrees at some large roll-tilt angles. These findings raise questions about the appropriateness of comparing the results of studies using the different tasks VPV and VPH. Factors that might contribute to this effect are discussed, including somatosensory input and ocular counterrolling (OCR).