The invention discloses an information fusion method in intelligent transformer station fault diagnosis. According to a protection configuration of each component in a system, a component-oriented Bayes network fault diagnosis model is built. Reliable parameters of a protector and a breaker are used as basic inputs of the model, and the model decides trust values among nodes of the model. The reliable parameters are defined on the basis of secondary network warning information. According to configuration information of an intelligent transformer station, a device - warning information relation matrix is set up. An error back propagation algorithm is used for training network parameters of the model. Fault warning information obtained during fault is used as input of a trained fault network. The value of a target node is calculated so that a fault probability of the component is obtained. The method solves the defect of using protection and breaker information only during primary system diagnosis. The method widens information sources and integrates information from multiple sources, so that the precision and reliability of diagnosis are improved.