Music performance requires that musicians represent many different kinds of sequence structure: musicians must remember which pitch to produce, when to produce it, and how to produce it (with what movements). The time course of item retrieval and movement preparation processes during music performance are considered. Serially ordered stage models of retrieval, in which item retrieval ends before movement preparation begins, are compared with interactive cascade models, in which the time course of both processes overlap, permitting interaction. Evidence from transfer of learning paradigms, production errors, and anticipatory movements, as measured in motion capture, are described. This early evidence suggests different time courses for item retrieval (slower, earlier) than for movement preparation (faster, later) with significant temporal overlap during music performance.