This paper presents a methodology for the control of a robot arm, using electromyographic (EMG) signals. EMG signals from the muscles of the shoulder and elbow joints are used to predict the corresponding joint angles and the force exerted by the user to the environment through his/her forearm. The user's motion is restricted to a plane. An analysis of various parametric models is carried out in order to define the appropriate form of the model to be used for the EMG-based estimates of the motion and force exerted by the user. A multi-input multi-output (MIMO) black-box state-space model is found to be the most accurate and is used to predict the joint angles and the force exerted during motion, in high frequency. A position tracking system is used to track the shoulder and elbow joint angles in low frequency to avoid drifting phenomena in the joints estimates. The high frequency model estimates, the low-frequency position tracker and a Kalman filter are used to control a torque controlled robot arm in the frequency of 500 Hz. The proposed system is tested both on teleoperation and orthosis scenarios. The experimental results prove the high accuracy of the system within a variety of motion profiles.