There are more than one motor unit activities recorded simultaneously during scanning EMG recordings. It is not possible to determine the phase duration correctly by inspecting only one sweep. The other motor unit activities, which are out of interest, should be filtered for this purpose. In this study, a new method is developed by revealing the activity corridor of scanning EMG recordings, wavelet transform based noise reduction, autocorrelation function based signal cutting location detection, and alpha trim filtering are applied to the recorded signals for determining phase duration correctly. Data are recorded from normal people and neuromuscular diseased people. Dataset contains 20 scanning EMG recordings.