Abstract: In order to diagnose faults for batch processes, a novel method, MPCA-MDPLS model, is presented in this paper. The proposed method includes two models: MPCA (multiway principal component analysis) model and MDPLS (multiway discriminant partial least squares) model. Based on data collected from the plant during normal operation and specific faults, two models are constructed.The MPCA model can detect unknown faults.The faults，which are detected by using MPCA model as not unknown,are further diagnosed by the MDPLS model. If it is identified as unknown, the root cause is analyzed by using various methods. The unknown fault is then saved in the historical database in order to reconstruct the MPCA-MDPLS model. The method is proved to be feasible and effective by the application in diagnosing a multi-stage streptomycin fermentation process.