We have evaluated the use of visual information about the movement of a target in two tasks--tracking and interceptions--involving multi-joint reaching movements with the arm. Target velocity was either varied in a pseudorandom order (random condition) or was kept constant (predictable condition) across trials. Response latency decreased as target velocity increased in each condition. A simple model that assumes that latency is the sum of two components--the time taken for target motion to be detected, and a fixed processing time--provides a good fit to the data. Results from a step-ramp experiment, in which the target stepped a small distance immediately preceding the onset of the ramp motion, were consistent with this model. The characteristics of the first 100 ms of the response depended on the amount of information about target motion available to the subject. In the tracking task with randomly varied target velocities, the initial changes in hand velocity were largely independent of target velocity. In contrast, when the velocity was predictable the initial hand velocity depended on target velocity. Analogously, the initial changes in the direction of hand motion in the interception task were independent of target velocity in the random condition, but depended on target velocity in the predictable condition. The time course for development of response dependence was estimated by controlling the amount of visual information about target velocity available to the subject before the onset of limb movement. The results suggest that when target velocity was random, hand movement started before visual motion processing was complete. The response was subsequently adjusted after target velocity was computed. Subjects displayed idiosyncratic strategies during the catch-up phase in the tracking task. The peak hand velocity depended on target velocity and was similar for all subjects. The time at which the peak occurred, in contrast, varied substantially among subjects. In the interception task the hand paths were straighter in the predictable than in the random condition. This appeared to be the result of making adjustments in movement direction in the former condition to correct for initially inappropriate responses.