This paper proposes an EMG-based music score input system for people with severe physical disabilities. The system discriminates EMG patterns measured from a user through a probabilistic neural network. The user can input music scores by selecting on operation commands using EMG signals displayed on a monitor. In this system, muscle activation levels and intervals between muscle contractions are allocated to control a cursor velocity and a musical note, respectively, thereby enabling efficient input of music scores. Muscle activation levels are also used to control sound volume for the input of espressivo music. Operation experiments involving a patient with cervical spine injury showed that the subject could control the system and input music using EMG signals measured from two pair of electrodes attached to the face. The system's suitability for the input of music score data was also validated by monitoring operation time intervals, the number of mistakes in operation and so on. The results indicated that the user could work with the system and freely input music score data.