The multiclassification algorithm st ructure of the support vector machine(SVM) has a decisive effect on the classifi cation ability. Based on Kernel function mapping, this paper presents an inter class separability method in high dimensional feature space called Jb ／Ｊｗ． Then the classifier s tructure of SVM is designed with the minmax rule. Finally, classifying experim ent of 4 types of rolling bearing faults is done, and the results show that the mult i fault classifier based on the J*b／J*w method and the minmax rule ca n identify the fault pattern more accurately and effectively.