Disclosed is a diagnostic method of the running smooth degree of a ball screw. The method comprises the steps that physical signal data generated by the ball screw are captured firstly, then according to the geometric dimension information and operation condition information of the ball screw, the physical signal data are partitioned to generate plural partitioned data, according to measurable dynamic characteristics of the ball screw in the running process, peak value signals of the partitioned data are quenched to generate a peak value characteristic data subsequence, plural peak value characteristic evaluation models and the peak value characteristic data subsequence are utilized for calculating a peak value characteristic data main sequence, and finally, according to the variability and distribution states of the data in the plural-peak-value characteristic data main sequence, plural smooth degree values are generated, and the smooth degree of running of the ball screw can be diagnosed through changes of the smooth degree values.