Traditional transformer fault diagnosis based on single source of information has significant limitation in identification of transformer fault type because of power transformer’s complex structure and changeable operating environment. So fusion technology is introduced into the fault diagnosis of power transformer in this paper. This method divides the progress of transformer fault diagnosis into two fusion levels. The first level is to ascertain whether it is overheated or discharged by content of gases dissolved in transformer oil. The second level is to ascertain the location or cause of the fault by electric data. The intelligence algorithm which is used in these two levels are both the improved BP neural network algorithm. Finally, the effectiveness is validated by the result of practical fault diagnosis examples.