Success in an imitation task may be defined in absolute terms (how well the reproduction matches the model) or in relative terms (how much better the reproduction matches the model than the imitator's spontaneous production matches the model). In the present study of expressive music performance, it was hypothesized that absolute imitation success would decrease but relative imitation success would increase as the model pattern diverges more and more from the typical pattern of spontaneous production, within reasonable limits. This hypothesis received support in Experiment 1, which required pianists first to play a musical excerpt spontaneously and then to imitate model performances instantiating different patterns of expressive timing or dynamics, with the other dimension held constant at a typical pattern. The typical expressive pattern for a given musical passage seems to function as a cognitive schema that biases perception, memory, and (re) production of expressive nuances. The results also suggested that imitation of different dynamic patterns affects expressive timing (louder tones were followed by longer interonset intervals), whereas imitation of different timing patterns had little effect on produced dynamics. The latter findings were replicated in Experiment 2, which presented models that differed in both timing and dynamics. Attention to both dimensions simultaneously also reduced imitation accuracy, especially for timing.