Abstract: Fault detection for multi-mode process is a complicated problem, as the fault detection for both steady mode and transition mode should be taken into consideration. Different modes are needed for different steady modes because different relations of variables are contained in each mode model. Transition mode is a dynamic process occurring when production changes operating mode. The dynamic characteristic reflects not only the changing variables but also the changing relation of variables. The relative change information can be obtained by differential transform of transition data. Principle components can be extracted by analyzing correlation of differential variables using principal component analysis (PCA). At last, segmented modeling with PCA method is used to monitor multi-mode process of continuous annealing line. The algorithm reflects transition process well and is proved to be efficient.