Abstract: An industrial boiler system is usually considered a single-mode system. When the system load changes, time-varying and multimodal characteristics are significantly exhibited, and the traditional PCA method has difficulty effectively monitoring faults. In view of this problem, we present a cross and piecewise PCA method to diagnose faults. The method selects a parameter value as the modal identification value of load variation, then multiple models are built for monitoring by means of cross and overlap segmentation. Crossover parts are defined as the mode transition processes, in which two models are used for comprehensive assessment to improve the accuracy of fault monitoring. The application in a practical industrial boiler shows that the method can significantly improve the accuracy of fault diagnosis compared with the traditional PCA method.