Four groups of 8 Ss each practiced a lever-positioning skill on 10 practice sessions. Group 1 received 10 2-min. trials of the complete task on each of the 10 sessions. The Ss of the other 3 groups practiced the whole performance on all trials of Sessions 1,9, and 10 and on trials 1 and 10 of all other practice sessions. Group 2 received four trials of practice on the horizontal component and four trials on the vertical component for Sessions 2 through 8. Group 3 received 8 trials of practice on the vertical component only and Group 4, 8 trials on the horizontal component only for Sessions 2 through 8. There was a clear superiority of whole practice over part practice. No reliable differences in performance were obtained between the 3 groups that received part practice. For all groups, however, there was significant acquisition over the practice sessions. All the learning curves showed significant departure from linearity. Performance on the components was slightly more than twice that on the whole task. The numerical superiority of Groups 3 and 4 on component practice over that for the subgroups of Group 2 is significant only for the vertical component. In the case of performance on the horizontal component, however, individual variability was about twice that of individual variability on the vertical component. In discussing the results, it is suggested that the superiority of whole practice may disappear at high levels of task complexity.