Abstract: An algorithm for recognizing peaks and data reduction in digitized power spectrum is detailed in this paper. It can be applied to extract line spectra features and reduce the amount of data in noise and vibration spectra. It is an efficient preprocessing means of the pattern recognitions which are dependent upon spectrum features. The kernel of this algorithm is the peak-recognition logic which is developed by fully considering the shape characteristics of spectra. In some practical applications, some good results have been obtained by this approach. The relevant application examples are given here.