Research has revealed that manual aiming movements performed with vision are more accurate than those performed without vision, and that removing visual information results in a large decrement in performance both early and late in practice. The dominance of visual feedback may result from participants using vision to adjust trajectories during movement execution (i.e., online) and/or from using visual feedback to enhance the programming of subsequent trials (i.e., offline). The objective of the present study was to determine what factors underlie the reliance of visual feedback for online and offline control. In Experiment 1, participants practiced a directional video aiming task under one of four feedback conditions: Full Vision, Delayed Trajectory Feedback, Endpoint Feedback, or Numeric Feedback. All participants were then transferred to a No-Vision condition without KR. The results indicated that the Full-Vision condition was more accurate than the other conditions during acquisition but suffered the largest decrement in performance during transfer. In Experiment 2, participants practiced the task under a Concurrent Vision condition, an Immediate Trajectory Feedback condition in which visual feedback of the trajectory was given upon movement completion, or a Delayed Trajectory Feedback condition in which visual feedback was presented 4 seconds after completion of the movement. Similar to Experiment 1, Concurrent Vision resulted in the highest accuracy levels during acquisition but the greatest decrement in performance when feedback was removed. In addition, the cost of removing feedback was greater in the Immediate Trajectory compared to the Delayed Trajectory condition. These results point to the reliance of visual feedback for both online and offline processing. The timing of visual feedback determines the degree to which participants rely on visual feedback for offline processing.