The invention relates to a fault diagnosis method for a rotating mechanical ball bearing based on K-means clustering and evidence reasoning. According to the method, a likelihood reliability table andK reference center vectors are obtained by K-means clustering and K reference evidences are obtained by the likelihood reliability table; distances between fault feature data and the K reference center vectors are calculated, the reference evidences are corrected, and K diagnostic evidences are generated; and after on-line acquisition of values of multiple kinds of fault characteristics, activated diagnostic evidences are calculated respectively, the activated diagnostic evidences are fused, a fault decision is made by using the fused evidences, and a fault type corresponding to the online fault characteristic data is obtained. According to the invention, the fusion reasoning of fault diagnosis evidences is performed based on K-means clustering; and with the multi-source diagnostic information, the fault diagnosis accuracy of the rotating mechanical ball bearing can be improved effectively.