Aiming at the classification of fault feature data for the nonlinear rotor system, a method is studied. On the premise of introduction of Biased Fisher Discriminatory Analysis, a method of dimension reduction for data set by KPCA combined with BFDA is introduced furthermore. The procedure of KPCA which removes redundant information of original data set is used to construct data set of dimension reduction, BFDA is used to reduce dimension of data set further and improve the separation of data subset from different category. The classification results of instance data and typical fault data show that，on the basis of dimensi on reduction separability of FDA, the proposed method of BFDA is lower for the characteristics of computational complexity. The result of dimension reduction corresponding to the proposed method of KPCA combined with BFDA can apply to linear classification directly, and effective data classification can be achieved.