This paper describes an experiment that aimed at examining the effects of extended practice in a sequential keypressing task. More specifically, predictions of three execution model were tested. First, the classic subprogram-retrieval model (Sternberg et al., 1978) assumes that an action sequence is programmed entirely before sequence execution while later, during sequence execution, instructions for the individual units in the sequence are retrieved. It predicts that keypressing rate decreases with sequence length but specific effects of practice are not predicted. The second mechanism, concurrent processing, assumes that various forms of information processing become increasingly concurrent in the course of practice. More specifically, it is proposed that with practice motor programming and retrieval of individual motor units from a short-term motor buffer concur with actual keypressing which, however, may delay execution of individual actions. Finally, associative motor unitization asserts that sequence execution comes to rely to an increasing degree on the automatic spread of activation among representations of the individual sequence units so that keypressing rate increases with position in the sequence. These and other predictions were verified with a two-choice sequence production task. One sequence contained four, the other two keypresses. The main results are: (a) a reduction with practice of the difference between the times to initiate long and short sequences, (b) practice had a larger effect on final keypresses than on earlier keypresses, and (c) practice had a larger effect on the final keypress of the long than that of the short sequence. An interesting additional effect was the occurrence of a pronounced `warming-up¿ effect for all interkey intervals at the second and third day of practice. The results were taken as evidence for the development of concurrent programming and concurrent retrieval while little support was observed for associative motor utinization and subprogram-retrieval.