Aimed at the defect of uncertainty of single sensor for the rolling bearing fault information recognition, the method of multisensor information fusion was proposed based on the BP neural network and the D-S evidence theory. Output results of BP neural network were normalized as the focal element of the basic probability assignment, five kinds of fault types of rolling bearing were identified as a system framework, and decision level fusion was made according to Dempster combination rule. The test results show that using the method in experiments of fault diagnosis for bearing inner ring wear, outer ring wear and ball bearing wear has improved the accuracy of fault diagnosis, and verified its feasibility.