The EMG signals which include information on not only mascle force but operator's intended motion and mechanical impedance property of joints have been often used as control signals of prosthetic arms. Most of previous researches, however, adopted only the ON/OFF control of the prosthetic arms depending on the results of the EMG pattern discrimination, or controlled only a particular joint depending on the torque estimated from the EMG signals. In this paper, we propose a multi-joint control method of a prosthetic forearm using the EMG signals. In order to realize the natural feeling of control similar to that of the human movements, the impedance model of human forearm is introduced to the control system. Also the force level during the motion is estimated from the EMG signals and used as the control command to each joint. It is shown from the experiments that the forearm motions and force levels can be estimated by using the EMG signals, and the prosthetic forearm can be controlled based on the impedance model smoothly.