1. Experiments have been conducted on human subjects to determine the role of prediction in smooth eye movement control. Subjects were required to actively pursue a small target or stare passively at a larger display as it moved in the horizontal plane. 2. Target motion was basically periodic, but, after a random number of cycles an unexpected change was made in the amplitude, direction or frequency of target motion. Initially, the periodic stimulus took the form of a square waveform. In subsequent experiments, a triangular or sawtooth waveform was used, but in order to examine the timing of the response in relation to stimulus appearance, the target was tachistoscopically illuminated for 40-320 ms at the time that it passed through the mid-line position. 3. When subjects either actively pursued the target or stared passively at the larger display a characteristic pattern of steady-state eye movement was evoked composed of two phases, an initial build-up of eye velocity that reached a peak after 200 ms, followed by a decay phase with a time constant of 0.5-2 s. The build-up phase was initiated prior to target displacement for square-wave motion and before onset of target illumination for other waveforms. 4. The peak eye velocity evoked gradually increased over the first two to four cycles of repeated stimulation. Simultaneously, the response became more phase advanced, the reaction time between stimulus onset and the time at which peak velocity occurred decreasing from an average of 300 to 200 ms for triangular waveform stimuli. 5. When there was a sudden and unexpected change in amplitude and direction of the stimulus waveform, the eye movement induced had a peak velocity and direction that was inappropriate for the current visual stimulus, but which was highly correlated with the features of the preceding sequence in the stimulus. 6. When there was a sudden change in the frequency of the stimulus waveform the predictive eye movement was induced with a timing appropriate to the periodicity of the previous sequence but inappropriate to the new sequence. 7. The results indicate that prediction is carried out through the storage of information about both the magnitude and timing of eye velocity. The trajectory of the averaged eye velocity response was similar in form irrespective of the duration of target exposure or basic stimulus frequency, suggesting that the predictive estimate is released as a stereotyped volley of constant duration but varying magnitude under the control of a periodicity estimator.(ABSTRACT TRUNCATED AT 400 WORDS)