Pursuit eye movements keep the image of a small moving target near the fovea with an admixture of smooth pursuit and saccades. To probe the initiation of pursuit, four humans and one monkey attended a bright laser spot that was rear-projected on a diffusely illuminated (1 cd/m(2)) tangent screen. Movement of the viewing eye was recorded with a magnetic search coil. After presenting a stationary target for a random time (1-3 sec), the computer extinguished the target in separate trial blocks for 40, 100, 200, or 500 ms. At the end of this gap period, the target reappeared on the fovea, moving centrifugally at 10°/sec in a random direction before decelerating to a stop. The predominate initial response (< 90%) to unpredictable changes in target motions had four sequential components: a latent period, a primary smooth movement, a small saccade, and then a secondary smooth movement. None of five measured variables depended systematically on the gap duration: the latency to primary smooth pursuit (range of means 144-217 ms), average eye velocity in the first 50 msec of primary smooth pursuit (0.7-3.6°/s), latency to the first saccade (207-319 ms), amplitude of the first saccade (1.6-2.7°), or average eye velocity in the first 100 ms of the secondary smooth pursuit (7.4-11-4°/s). Our data do not support the hypothesis that a separate fixation system has to be disengaged before smooth pursuit can begin, as has been suggested for the saccadic system.