In the industrial production process containing a number of different modes, there is a certain transition process between every two steady modes. Although the transition period is rather short, its the complex dynamic behavior makes traditional process monitoring methods difficult to obtain satisfactory results. Therefore, a process monitoring method based on multi-mode identification is proposed. Firstly, the identification of the basic steady modes is realized through a window cutting method. Then, the accurate time boundaries of transitional modes are determined by using a moving window strategy, and a differential segmentation technique is performed on the sub-modes of the transitional process for classification. Considering different distribution characteristics of each data segment, independent component analysis and principal component analysis are carried out to deal with the non-Gaussian and Gaussian information. Finally, according to the Bayesian inference, three statistics are reconstructed to realize the multi-mode process monitoring. The feasibility and effectiveness of the proposed method are later demonstrated through a simulated Tennessee Eastman(TE) process.