1. To evaluate the role of the cerebellum during improvement of voluntary motor performance over time, the discharge of 88 Purkinje cells in the intermediate and lateral cerebellum of two primates (Macaca mulatta) was investigated during a motor learning task involving visually guided arm movements. The animals were trained to move a draftsman's style manipulandum over a horizontally placed video screen. The animals were required to move a cursor from the start box to one of four target boxes by movement of the manipulandum. Errors were introduced into the movement by altering the visual feedback loop, changing the gain between the cursor movement and the hand movement. When a novel pin was presented over 100-200 movement trials, the animals adapted the movements to the new gain. The animals used a strategy of scaling the amplitude and velocity of the initial phase of the movement while keeping the time to peak velocity constant. 2. The learning paradigm consisted of an initial control phase with 35-100 trials at the gain of 1.0. The next 100-200 trials, the learning phase, were presented at one of four gains (0.6, 0.75, 1.5, 2.0). Lastly, a testing phase involved 80% of 100 trials at the learned pin and 20% of the trials randomly interspersed at the control gain of 1.0. An additional ''distance control'' was used in most experiments to control for the movement scaling associated with learning. In this series of movements using a gain of 1.0, the target box was placed at the distance and direction the hand would have to move in the adapted state. Two aspects of the kinematics were the same for the distance control and the movement at the learned pin: movement amplitude and time to peak velocity. There were, however, slight differences in the peak velocity attained. For gains < 1.0, the peak velocity of the learned task was 14-20% lower than the distance controls, and for gains > 1.0, it was 10-18% higher. 3. After implantation of chronic unit recording hardware, Purkinje cell simple and complex spike discharge was recorded extracellularly during the learning task. The cells were located primarily in the ipsilateral intermediate zone or nearby hemisphere of lobules V and VI. Simple and complex spike histograms, as well as averages of the hand displacement and velocity profiles, were calculated for each phase of the paradigm. To determine the time course of any changes, the learning trials were subdivided into three equal phases. Spike discharge and kinematic averages were also calculated for each of the three phases. Simple and complex spike responses, as well as simple spike background activity, were statistically analyzed for changes during learning. 4. Statistically significant but transient changes occurred in complex spike discharge. During the learning paradigm, 43% (38/88) of the Purkinje cells exhibited a statistically significant change in complex spike activity. For 76% (29/38), the changes were transient, the complex spike discharge returning to control levels in the testing phase. The changes in complex spike responses were before or during the early phase of the movement (average 35 ms before movement onset to 238 ms after). For cells with a transient increase in complex spike discharge, the phase of maximum activity was almost equally likely to occur during the first, second, or third subperiod of the total learning phase. 5. Using the distance control to account for the different amplitude movements before and after learning, the simple spike response changed during the movement in 34% (22/65) of Purkinje cells in a manner that could not be attributed to alterations in movement distance. Of the 31 cells with a significant change in complex spike discharge during learning in which a distance control was obtained, 10 (32%) exhibited this type of simple spike change. However, of the 34 Purkinje cells with no significant complex spike change, the simple spike response was altered in 12 (35%). The magnitude of the changes in the simple spike responses was not statistically different in the two groups, nor were the response onset times, recording locations, or receptive field properties. 6. Simple spike background firing rate changed during the learning paradigm. In 65% (42/65) of the Purkinje cells studied, progressive changes occurred in the background simple spike discharge that could not be accounted for by movement distance. Nearly equal numbers of cells exhibited increasing (n = 18) and decreasing (n = 24) simple spike background activity. When grouped according to whether complex spike modulation changed during learning, the changes in background firing rate were not significantly different in these two groups. 7. The timing of the complex spike and simple spike changes were examined in relation to the learning process using total movement duration as a measure of the adaptation. Adaptation occurred in different subperiods of the learning trials. The majority of complex spike and simple spike responses during movement and simple spike background changes occurred before or in the same phase in which total movement time stabilized. A similar analysis based on the scaling of peak velocity yielded comparable results. 8. From these data, we conclude that the climbing fiber system is transiently activated during the process of learning to scale arm movements. One of the predicted consequences of this complex spike activity based on the Marr-Albus hypothesis, long-term alterations in simple spike discharge, is not supported by the data. Both changes in the movement-related simple spike modulation and background activity were as likely to occur with or without complex spike activation during learning and were of comparable magnitude in the two groups. However, the changes in complex spike and simple spike activity imply a role for the cerebellar cortex in the refinement of motor performance characteristic of learning.