We describe in detail the frequency response of the human three-dimensional angular vestibulo-ocular response (3-D aVOR) over a frequency range of 0.05-1 Hz. Gain and phase of the human aVOR were determined for passive head rotations in the dark, with the rotation axis either aligned with or perpendicular to the direction of gravity (earth-vertical or earth-horizontal). In the latter case, the oscillations dynamically stimulated both the otolith organs and the semi-circular canals. We conducted experiments in pitch and yaw, and compared the results with previously-published roll data. Regardless of the axis of rotation and the orientation of the subject, the gain in aVOR increased with frequency to about 0.3 Hz, and was approximately constant from 0.3 to 1 Hz. The aVOR gain during pitch and yaw rotations was larger than during roll rotations. Otolith and canal cues combined differently depending upon the axis of rotation: for torsional and pitch rotations, aVOR gain was higher with otolith input; for yaw rotations the aVOR was not affected by otolith stimulation. There was a phase lead in all three dimensions for frequencies below 0.3 Hz when only the canals were stimulated. For roll and pitch rotations this phase lead vanished with dynamic otolith stimulation. In contrast, the horizontal phase showed no improvement with additional otolith input during yaw rotations. The lack of a significant otolith contribution to the yaw aVOR was observed when subjects were supine, prone or lying on their sides. Our results confirm studies with less-natural stimuli (off-vertical axis rotation) that the otoliths contribute a head-rotation signal to the aVOR. However, the magnitude of the contribution depends on the axis of rotation, with the gain in otolith-canal cross-coupling being smallest for yaw axis rotations. This could be because, in humans, typical yaw head movements will stimulate the otoliths to a much lesser extent then typical pitch and roll head movements.