A novel approach to harmonic analysis for industrial power systems based on windowed fast Fourier transform is presented. First, the desirable sidelobe window (DSW) is constructed through self-convolution of the maximum decay sidelobe window (MDW) in the time domain. Then, the power system signal parameters, i.e., frequency, phase, and amplitude, are calculated by the DSW-based discrete phase difference correction algorithm. It has been shown that the spectral leakage and harmonic interferences can be reduced considerably by weighting samples with the DSW because of its good sidelobe behaviors, i.e., the peak sidelobe level and the sidelobe rolloff rate are proportional to the order of the window (or the times of self-convolution). The DSW-based discrete phase difference correction algorithm can be easily implemented in embedded systems due to the low computation burden. Both simulative and experiment data tests have been performed to verify the effectiveness and practicability of the proposed method.