1. We examined the relationship of cerebellar Purkinje cell discharge to the scaling of kinematics during a voluntary motor learning paradigm. The study focused on whether the occurrence of complex spike (CS) discharge was associated with kinematic changes. Two primates (Macaca mulatta) were trained to move a cursor using a two-joint manipulandum over a horizontal video screen from a start target to one of four target boxes. The relationship between the cursor and the hand (gain) was changed, requiring scaling of movement distance to complete the task. As previously described, when the novel gain was presented over 100-200 movement trials the animals adapted their movements by using a strategy of scaling the amplitude and velocity of the first phase of the movement while keeping time to peak velocity constant. 2. The paradigm consisted of four different phases. A control phase at a gain of 1.0 was initially performed. The learning phase over the next 180-210 movements used one of four gains (0.6, 0.75, 1.5, or 2.0). Last, a testing phase involved 80% of 100 trials at the learned gain and 20% of the trials at the control gain of 1.0. The distance control phase consisted of using a gain of 1.0 but having the animal move to targets placed at the distance and direction the hand moved in the adapted state. 3. Simple spikes (SSs) and CSs of 141 Purkinje cells recorded primarily in the intermediate and lateral regions of zones V and VI in three cerebellar hemispheres from the two primates were recorded during the distance control, control, learning, and testing phases. Some cells were recorded in lobule VII and Crus I. CS activity increased during the learning phase, as documented previously. The increase in CS discharge occurred before or during the first 200-300 ms of the movement. This is the same time period in which the kinematic changes necessary for adaptation to the novel gain occur. Of 141 Purkinje cells recorded during the learning paradigm, 104 (74%) demonstrated significant increases in CS firing rate during the learning-testing phase. Of these 104 cells, 82 had statistically significant SS modulation. 4. Movement trials with CSs were separated from the trials without CSs. Aligning the kinematic and spike train data on movement onset, the average velocity profiles were subtracted from each other and a strict statistical criterion applied to test for the significance of any differences. Movement trials randomly sorted into two groups served as a control. The kinematic and spike train data were also sorted into two groups but aligned on the occurrence of CSs. A Monte Carlo- type procedure was used to simulate CS times for the non-CS movement trials. Of the 104 Purkinje cells with a CS response, in 81 (78%) velocity was significantly different during the movements with CSs compared with the movements without CSs. This was observed whether the movements were aligned on movement onset or CS occurrence. 5. The CS-associated velocity changes were uni- or bimodal in shape. Mean maximum velocity difference was comparable in magnitude whether aligned on movement onset or CS occurrence. However, time of peak velocity difference was more closely associated with CS occurrence than with movement onset. Likewise, the latency of the velocity change was tightly coupled to CS occurrence. Of the 35 Purkinje cells with significant bimodal CS-aligned differences during CS trials, the mean latency of the first velocity peak preceded the CS response onset by 71 ms, whereas the mean latency of the second peak was 205 ms after the CS response. 6. The absolute amplitude of the velocity change in the CS trials was related to the feedback gain, with the smaller gains (i.e., adaptation requires larger movement amplitudes) associated with larger velocity differences. A contingency analysis showed that the relationship between the feedback gain of the learning series and the velocity difference is a dependent one. When adapting to the high gains, the CSs occur preferentially in movements with a larger velocity. Furthermore, for those cells in which the CS occurrence is associated with a decrease in velocity, the feedback gain had a greater probability of being <1. This finding suggests that CSs are more likely to occur in trials in which the velocity difference is inappropriate for the feedback gain the animal is required to learn.