The amount and quality of practice predicts expertise, yet optimal conditions of practice have primarily been explored with novice learners. Ten expert musicians and ten novices practiced disc-throwing skills under self-regulated conditions. A third novice group practiced with the same schedule as the music experts (yoked). The groups did not differ in terms of the amount of contextual interference, only in terms of when in-practice interference was introduced. The music experts progressed from a more blocked to random schedule which was opposite to the novices. This resulted in more accurate performance in retention for the experts in comparison to both novice groups (self-scheduled and yoked). The music expert and yoked groups showed higher form scores than the novice self-scheduled group, which might be related to the greater frequency of augmented information for these groups. There was no evidence that non-task-domain experts choose a more random practice schedule than novices, but in accord with good practice principles, they gradually introduced high amounts of interference into their practice. This strategy was associated with less error in retention for the experts. Because the yoked group showed more error than the music experts, the advantage of this schedule was also performance dependent. (Contains 3 figures and 1 table.)