1. Three monkeys were trained to make whole-arm reaching movements from a common central starting position toward eight radially arranged targets disposed at 45 degrees intervals. A sample of 312 cerebellar neurons with proximal-arm receptive fields or discharge related to shoulder or elbow movements was studied in the task. The sample included 69 Purkinje cells, 115 unidentified cortical cells, 65 interpositus neurons, and 63 dentate units. 2. The reaching task was divided into three movement-related epochs: a reaction time, a movement time, and holding over the target. All neurons demonstrated significant changes in discharge during one or more of these three epochs. Almost all of the cells (95%) showed a significant change in activity during the movement, whereas 68-69% of the cells showed significant changes from premovement activity during the reaction time and holding periods. 3. During the combined reaction time-movement period, 231/312 cells were strongly active in the task. Of these, 151 cells (65.4%) demonstrated unimodal directional responses. Sixty-three had a reciprocal relation to movement direction, whereas 88 showed only graded increases or decreases in activity. A further 37 cells (16.0%) were nondirectional, with statistically uniform changes in discharge in all eight directions. The remaining 43 cells (18.6%) showed significant differences in activity for different directions of movement, but their response patterns were not readily classifiable. 4. The proportion of directional versus nondirectional cells was consistent across the four cell populations. However, graded response patterns were more common and reciprocal responses less common among Purkinje and dentate neurons than among unidentified cortical cells and interpositus neurons. 5. The distribution of preferred directions of the population of cerebellar neurons covered all possible movement directions away from the common central starting position in the horizontal plane. When the preferred direction of each cell in the sample population was aligned, the mean direction-related activity of the cerebellar population formed a bell-shaped tuning curve for the activity recorded during both the reaction time and the movement, as well as during the time the arm maintained a fixed posture over the targets. A vector representation also showed that the overall activity of the cerebellar population during normal reaching arm movements generated a signal that varied with movement direction. 6. These results demonstrate that the cerebellum generates a signal that varies with the direction of movement of the proximal arm during normal aimed reaching movements and is consistent with a role in the control of the activity of muscles or muscle groups generating these movements.