To investigate whether the primary planes of eye and body responses to galvanic vestibular stimulation (GVS) are congruent, we have measured the binocular, three-dimensional eye movements (scleral coil technique) to bilateral bipolar GVS in six normal human subjects. Stimulation intensities were kept deliberately low in order to characterize the response to near-threshold intensities of stimulation (0.1-0.9 mA) that had been used previously to characterise body postural responses. Stimuli were applied for 4 s, but only the early responses that occurred within the initial 300 ms of turning the current on or off were measured. At intensities of 0.1-0.7 mA the 'on' response consisted almost exclusively of a torsional slow phase eye movement in which the top of the eyes rotated towards the anode. The latency of the torsional response was ca. 46 ms. A weak polarity-dependent disconjugate response was also observed in which the intorting eye elevated and the extorting eye depressed ('skew eye deviation'). When the current was turned off similar responses occurred in the reverse direction. Removal of the visual fixation light-emitting diode (LED) had no consistent effect on the short-latency ocular responses. The direction of the ocular response was similar to that of the postural response and is compatible with GVS inducing an apparent dynamic roll-tilt of the head towards the cathode. However, weak horizontal eye movements, which became more prominent as the stimulus intensity was increased to 0.9 mA, were also observed. This suggests that an additional weak rotational component about the yaw axis, or a component of lateral translation in the frontal plane, is contained in the GVS-evoked signal. The overall pattern of eye movement suggests that semicircular canal afferents contribute to these low-intensity GVS responses.