When people repeatedly practice the same cognitive task, their response times (RT) invariably decrease. Dutilh, Vandekerckhove, Tuerlinckx, and Wagenmakers (2009) argued that the traditional focus on how mean RT decreases with practice offers limited insight; their diffusion model analysis showed that the effect of practice is multifaceted, involving an increase in rate of information processing, a decrease in response caution, adjusted response bias, and, unexpectedly, a strong decrease in nondecision time. In this study, we aim to further disentangle these effects into stimulus-specific and task-related components. The data of a transfer experiment, in which repeatedly presented sets and new sets of stimuli were alternated, show that the practice effects on both speed of information processing and time needed for peripheral processing are partly task-related and partly stimulus-specific. The effects on response caution and response bias appear to be task-related. This diffusion model decomposition provides a perspective on practice that is more detailed and more informative than the traditional analysis of mean RT.