D-S evidence theory is an effective method of processing uncertain information. It is widely used in data fusion systems. However, there are two difficulties in the application: evidence conflict and requested computing time of Dempsterpsilas arithmetic that is power exponential increasing relation with number of data sources. Starting with the basic conception of D-S evidence theory, we generalize the rule for combining pieces of evidence, find the way to solve the two difficulties, build multi-source data fusion fault diagnosis system based on D-S evidence theory. Results of examples show that the system uses redundant complementary information of multi-source data, it is good to improve the reliability and exact ratio of fault diagnosis, and decrease the uncertainty.