Abstract: To improve the nonlinear and non-stationary characteristics of rolling bearing vibration signal, a multifractal descending algorithm was proposed to calculate the multifractal spectrum parameters. Taking the multifractal spectrum parameters as characteristic parameters, the advantages and disadvantages of multifractal descending fluctuation analysis method and multifractal descending moving average method were compared and analyzed for bearing fault feature extraction. An improved K mean clustering analysis was used to classify the feature parameters extracted from the multifractal descending algorithm, so as to realize the purpose of bearing fault diagnosis. The rolling bearing data were used to verify the proposed method, and the time domain characteristics and multifractal spectrum parameters were compared and analyzed. And the effects of two multifractal descending algorithms were compared and analyzed. The results verify the combination effectiveness of multifractal descending wave analysis and improved K means clustering for bearing fault diagnosis, and provide a new attempt for bearing fault diagnosis.