This paper presents an electromyographic (EMG)-based continuous control scheme including simple classifier for an electric-powered wheelchair, ultimately for quadriplegics. The proposed scheme utilizes three EMG signals as inputs for the muscle-computer interface. Since zygomaticus major muscles and transversus menti muscle of human face are able to move independently as well as to adjust contractile forces voluntarily, the surface EMG signals on these muscles are utilized for the electric-powered wheelchair control system. To extract the envelopes of the signal waveforms and to reflect the moving average activities, the root-mean-squares (RMS) operation and normalization are subsequently employed as initial signal processing. Then, an activation vector containing three normalized RMS signals is obtained in real time. The activation vector is applied to the simple classifier for finding out the motion command. Both desired linear acceleration and angular velocity are yielded from the linear combinations of the classification result and the magnitude of activation vector. Finally, desired wheel velocities of the wheelchair control system are obtained by using the integration and differential inverse kinematics. The effectiveness of the proposed control scheme is verified through several experiments such as avoiding obstacle cones and navigating long distance by the users.