There have been many algorithm proposed to solve music classification problems. The composition of music context is complicated, and the music genre is defined by musical perception. Lacking of qualifications to determinate music genres makes music classification more difficult. In this paper, method based on genetic expression programming to classify Midi music files was proposed. This method uses statistical information of Midi file features to classify Midi music genres, and builds models and classification rules. The result can be use for music recommendation or classification systems.