We report five experiments in which listeners heard the beginnings of classical minuets (or similar dances). The phrase in either measures 1-2 or measures 3-4 was selected as a target, tested at the end of the excerpt. A "beep" indicated the test item, which was a continuation of the minuet as written. Test items were targets (repetitions of the selected phrase), similar lures (imitations of targets), or different lures , and occurred after delays of 4-5, 15, or 30 s. We estimated the proportion of correct discriminations of targets from similar lures and targets from different lures. In Experiment 1, discrimination of targets from similar lures (but not of targets from different lures) improved between 5 and 15 s. Experiment 2 extended this result to a delay of 30 s. Discrimination of targets from similar lures improved over time, especially for second-phrase targets. This improvement was due mainly to decreasing false alarms to similar lures. Experiments 3 and 4 replaced the continuous music with silence and with a repetitive "oom-pah-pah" pattern, and the improvement in discrimination of targets from similar lures disappeared. Experiment 5 removed listeners' expectations of being tested, and the improvement also disappeared. Results are considered in the framework of current theories of memory, and their implications for the listener's experience of hearing music are discussed.