ElectroMyoGraphy (EMG) signal decomposition is the inverse process of the generation of EMG signals. The complete EMG decomposition was completed based on the superposition waveforms resolution in order to obtain information about Motor Unit (MU) template waveform and firing pattern. Firstly, noise was removed from the original EMG signals based on wavelet filtering and wavelet threshold estimation; then all the Motor Unit Action Potential (MUAP) waveforms were detected using amplitude-slope double threshold filtering, and all the detected MUAPs were classified into their constituent Motor Unit Action Potential Trains (MUAPT) through fuzzy K-means clustering and minimum distance classifier. Finally the superposition waveforms resolution procedure was finished using pseudo-correlation technique and peeling-off technique. This decomposition system has been evaluated using synthetic and real EMG signals. The average accuracy of the EMG decomposition system was above 87%.