1. We studied the contribution of the individual semicircular canals to the generation of horizontal and torsional eye movements in cynomolgus monkeys. Eye movements were elicited by sinusoidal rotation about a vertical (gravitational) axis at 0.2 Hz with the animals tilted in various attitudes of static forward or backward pitch. The gains of the horizontal and torsional components of the vestibuloocular reflex (VOR) were measured for each tilt position. The gains as a function of tilt position were fit with sinusoidal functions, and spatial gains and phases were determined. After control responses were recorded, the semicircular canals were plugged, animals were allowed to adapt, and the test procedure was repeated. Animals were prepared with only the anterior and posterior canals intact [vertical canal (VC) animals], with only the lateral canals intact [lateral canal (LC) animal], and with only one anterior and the contralateral posterior canals intact [right anterior and left posterior canal (RALP) animals; left anterior and right posterior canal (LARP) animals]. 2. In normal animals, the gain of the horizontal (yaw axis) velocity of the compensatory eye movements decreased as they were pitched forward or backward, and a torsional velocity appeared, reversing phase at the peak of the horizontal gain. After the anterior and posterior canals were plugged (LC animal), the horizontal component was reduced when the animal was tilted backward; the gain was zero with about -60 degrees of backward tilt. The spatial phase of the torsional component had the same characteristics. This is consistent with the fact that both responses were produced by the lateral canals, which from our results are tilted between 28 and 39 degrees above the horizontal stereotaxic plane. 3. After both lateral canals were plugged (VC animals), horizontal velocity was reduced in the upright position but increased as the animals were pitched backward relative to the axis of rotation. Torsional velocities, which were zero in the upright position in the normal animal, were now 180 degrees out of phase with the horizontal velocity. The peak values of the horizontal and torsional components were significantly shifted from the normal data and were closely aligned with each other, reaching peak values at approximately -56 degrees pitched back (-53 degrees horizontal, -58 degrees torsional). The same was true for the LARP and RALP animals; the peak values were at -59 degrees pitched back (-55 degrees horizontal, -62 degrees torsional). Likewise, in the LC animal the peak yaw and roll gains occurred at about the same angle of forward tilt, 35 degrees (30 degrees horizontal, 39 degrees torsional). Thus, in each case, the canal plugging had transformed the VOR from a compensatory to a direction-fixed response with regard to the head. Therefore there was no adaptation of the response planes of the individual canals after plugging. 4. The data were compared with eye velocity predictions of a model based on the geometric organization of the canals and their relation to a head coordinate frame. The model used the normal to the canal planes to form a nonorthogonal coordinate basis for representing eye velocity. An analysis of variance was used to define the goodness of fit of model predictions to the data. Model predictions and experimental data agreed closely for both normal animals and for the animals with canal lesions. Moreover, if horizontal and roll components from the LC and VC animals were combined, the summation overlay the response of the normal monkeys and the predictions of the model. In addition, a combination of the RALP and LARP animals predicted the response of the lateral-canal-plugged (VC) animals. 5. When operated animals were tested in light, the gains, peak values, and spatial phases of horizontal and roll eye velocity returned to the preoperative values, regardless of the type of surgery performed. This indicates that vision compensated for the lack o