Eye movements were recorded in human subjects who tracked a target spot which moved horizontally at constant speeds. At random times during its trajectory, the target disappeared for variable periods of time and the subjects attempted to continue tracking the invisible target. The smooth pursuit component of their eye movements was isolated and averaged. About 190 ms after the target disappeared, the smooth pursuit velocity began to decelerate rapidly. The time course of this deceleration was similar to that in response to a visible target whose velocity decreased suddenly. After a deceleration lasting about 280 ms, the velocity stabilized at a new, reduced level which we call the residual velocity. The residual velocity remained more or less constant or declined only slowly even when the target remained invisible for 4 s. When the same target velocity was used in all trials of an experiment, the subjects' residual velocity amounted to 60% of their normal pursuit velocity. When the velocity was varied randomly from trial to trial, the residual velocity was smaller; for target velocities of 5, 10, and 20 deg/s it reached 55, 47, and 39% respectively. The subjects needed to see targets of unforeseeable velocity for no more than 300 ms in order to develop a residual velocity that was characteristic of the given target velocity. When a target of unknown velocity disappeared at the very moment the subject expected it to start, a smooth movement developed nonetheless and reached within 300 ms a peak velocity of 5 deg/s which was independent of the actual target velocity and reflected a "default" value for the pursuit system. Thereafter the eyes decelerated briefly and then continued with a constant or slightly decreasing velocity of 2-4 deg/s until the target reappeared. Even when the subjects saw no moving target during an experiment, they could produce a smooth movement in the dark and could grade its velocity as a function of that of an imagined target. We suggest that the residual velocity reflects a first order prediction of target movement which is attenuated by a variable gain element. When subjects are pursuing a visible target, the gain of this element is close to unity. When the target disappears but continued tracking is attempted, the gain is reduced to a value between 0.4 and 0.6.