In this project, we created an agent-based model of music group playing under four di↵erent interaction mechanisms. Based on real music data, added randomness and simplifying assumptions, we examine how agents synchronize and deviate from the original score. We find that while music can make synchronization complex, it also helps reducing the total deviation. By studying the simulation process, several conclusions on the relationship between di↵erent growing speeds of total deviations and di↵erent interaction schemes are drawn. With interpretation from a musical point of view, we find that, in a music ensemble, listening to neighbors helps the players end up in sync. However, if people do not listen carefully enough, the deviation becomes larger than when people do not listen at all. On the issue of whom one should listen to, the results show no significant di↵erences between listening to the immediate neighbors and to the whole group. Finally, we also observe that large deviations can be reduced by making the musicians move while playing.