1. We studied eye velocity during the first 2 s of the vertical vestibuloocular reflex (VOR) elicited from cats placed on their sides (90° roll position) and rotated about an earth vertical axis. Vestibular stimuli were presented in the dark and consisted of brief trapezoidal velocity profiles. Eye movements were recorded with a magnetic search coil, and eye velocity was analyzed with high temporal resolution. 2. The first 2 s of upward or downward eye velocity after the onset of head rotation was characterized and compared. Adaptive changes in VOR gain (eye/head velocity) were then induced, and upward and downward eye velocity responses were again compared. 3. The early time course of the vertical VOR was complex. After a latency of ~15 ms, eye velocity increased rapidly until it was equal in magnitude and opposite in direction to head velocity. The peak eye velocity decayed within <1 s to a plateau of slow-phase eye velocity (SPEV) equal to ~-0.6 times the head velocity. Peak upward and downward eye velocity was symmetric. The transition from peak to plateau was more rapid for the downward VOR (slow phases downward) than for the upward VOR (slow phases upward). The plateau attained by upward SPEV was ~15% higher than the plateau attained by downward SPEV. 4. VOR gain adaptation was symmetric. The percentage change in adapted upward eye velocity equalled the percentage change in adapted downward eye velocity. Both peak and plateau SPEV adapted, but peak eye velocity adapted less than plateau eye velocity. VOR latency was unchanged by adaptation. 5. The trajectory of the VOR response to steps of head velocity could be divided into an invariant and a variant interval. The invariant interval consisted of the initial ~15 ms of the eye movement. Neither direction of head movement (upward vs. downward) nor adaptation of the VOR gain effected the eye movement trajectory during the invariant interval. The variant interval began ~30 ms after the onset of head movement and ~15 ms after the onset of eye movement. In unadapted animals, downward eye speed exceeded upward eye speed during the variant interval. In adapted animals, eye speed during the variant interval, but not during the invariant interval, diverged from eye speed in the unadapted state. We suggest that the initial invariant interval (~15 ms) of the eye movement response trajectory may represent the direct response of the classically described three-neuron arc. The variant interval may represent a contribution of longer latency pathways mediating VOR gain adaptation.