1. We have tested the hypothesis that the basal ganglia initiate some one or several modes of movement by recording the change in discharge frequency of pallidal neurons during visually triggered step and visually paced ramp moves in relation to the visual stimulus onset, the change in the electromyograph (EMG), and the movement onset of trained rhesus monkeys. 2. The modal times of change for globus pallidus pars interna (GPi) were significantly later than those for forearm agonist muscle EMG. By contrast, the modal time of change for the cerebellar dentate nucleus preceded that for wrist agonist EMG. 3. The direction of change in discharge frequency of the GPi cells was for 71% an increase and for 29% a decrease. 4. Because of the relatively late change of activity of GPi neurons, we propose that GPi neurons cannot initiate these movements, as others have also suggested. The commands for the initiation of these movements may instead be generated by structures that include the lateral cerebellum and the anterior cerebral cortex. 5. We have also tested the hypothesis that the pallidum of the basal ganglia or the dentate of the lateral cerebellum may control the direction and other parameters of the trajectory by recording from both structures to see whether cell discharge correlated with the parameter and whether the correlation was consistent across tasks. Two rhesus monkeys were trained to perform hold-ramp-hold and hold-step-hold visually guided movements in opposite directions by flexing and extending the wrist with and against uniform oppositely directed torque loads (0.2 Nm). Wrist position, velocity, force, and EMG were recorded simultaneously. Movement amplitudes and directional intent were computed and inferred, respectively. 6. Task related neurons were classified as follows: 1) directional, if the discharge rate was reciprocal for opposite movements or if it increased or decreased during movement in one direction only; 2) bidirectional, if the discharge rate increased or decreased during movement in both directions; and 3) "other," if it was directional under one load and bidirectional under the other. During step tracking, 34 GPi, 47 globus pallidus pars externa (GPe), and 44 cerebellar dentate nuclear neurons were related to the task. Of the GPi cells, 14 (41%) were directional, 6 (18%) bidirectional, and 14 (41%) other. Of the GPe neurons, 13 (28%) were directional, 19 (40%) bidirectional, and 15 (32%) other. Of the dentate cerebellar nuclear cells, 5 (11%) were bidirectional, 31 (70%) bidirectional, and 8 (18%) other.(ABSTRACT TRUNCATED AT 400 WORDS)