Multi-way principal component analysis (MPCA) is an effective method for fault diagnosis in batch processes. In MPCA, the determination of principal component numbers(PCs) is the key to the model, which concerns the reliability, accuracy, completeness of PCA model. The traditional method, using CPV to determine PCs, is too subjective and does not consider the failure factors. In order to improve the detection performance, and effectively extract principal component, this paper proposes a method that is combing SNR with MPCA to select PCs in batch process, SNR indicates the relationship between the sensitivity of fault diagnosis and PCs. On the basis that the principal information fully describes the data ,and considering the influence of fault information on CPs, then it selects principal component. Applying this method to fault diagnosis in penicillin batch fermentation process, the simulation results show that the response curves of T² statistics and SPE statistics are more sensitive to fault, which effectively improves the accuracy rate of fault diagnosis.