1. The normal and adapted vestibuloocular reflex (VOR) of goldfish was characterized by means of sinusoidal, velocity step, and position step head rotations about the vertical axis. VOR adaptation was induced by short-term, 1- to 4-h, presentation of visual and vestibular stimuli that altered the ratio of eye to head velocity. 2. The VOR response measured with sinusoidal oscillations in the dark was close to ideal compensatory values over 2 decades (1/32-2 Hz). Gain approximated unity, and phase, in relation to the head, was nearly 180 degrees. The VOR was linear within the range of head velocity tested (4-64 degrees/s). 3. Head velocity steps from 1/8 to 1 Hz produced steplike eye velocity profiles that could be divided into an early acceleration-related "dynamic" component and a later constant-velocity "sustained" period frequently separated by a sag at approximately 0.1-0.15 s from the initiation of eye movement. The sustained response exhibited no decay during the constant-velocity component of the step. 4. Higher temporal resolution of the dynamic response showed the adducting eye movement to have a shorter latency, faster rise time, and larger peak gain than the abducting eye movement. The characteristics of this directional asymmetry were similar for position steps and electrical stimulation of the vestibular nerve. However, the asymmetry was not observed during sinusoidal head rotation, the sustained component of the step response, or after electrical stimulation of the VIth and IIIrd nerves. We conclude that this directional asymmetry is of central origin and may be largely due to the parallel vestibular and abducens internuclear neuron pathways onto medial rectus motoneurons. 5. The VOR adaptation process for both higher and lower eye velocity exhibited an exponential time course with time constants of 55 and 45 min, respectively. After continuous sinusoidal training for 4 h, VOR gain reached an asymptotic level 5% away from perfect suppression in the low-gain training, but 19% away from the actual performance in the high-gain paradigm. The time constant for VOR gain reversal was 5 h, and an asymptotic level 40% less than performance was reached within 10 h. 6. Adapted VOR gain was symmetrical for both directions of eye movement measured either during sinusoidal rotation or the sustained part of the velocity step. VOR adaptation also produced a comparable gain change in the nasal and temporal directions of the dynamic component, but this reflected the asymmetric characteristics observed in the preadapted condition.(ABSTRACT TRUNCATED AT 400 WORDS)