EMG has been extensively used to study function of back muscle, which plays an important role in the objective assessment of occupational low back pain, yet the inherent large variability of EMG signals across subjects produced by reasons already known or unknown may mask true biological differences. Some useful parameters abstracted from 8-channel EMG signal in time-frequency domain are used or proposed and calculated in this paper to decrease this variability, when comparing the possible difference between low back pain patients and normal control group. The calculation of instantaneous median frequency is improved to decrease the interruption of background noise, and different power densities in EMG time-frequency spectrogram are observed in two tested groups.