The traditional methods of system fault diagnosis are unable to automatically diagnose in the absence of historical statistics. In view of the above problem, a method of automatic fault diagnosis based on fussy FMEA analysis is proposed in this paper. By combining with fault modes, fault effect and diagnosis models of harm analysis, the method fulfills fault diagnosis based on different measured data. Fuzzy theory modeling is introduced while risk priorities number (RPN) is calculated. By means of gray correlation theory, fuzzy RPN (FRPN) of harm levels is concluded by solving fussy set, thus the expertise is merged into the evaluation model in the absence of statistics. Intelligent self-diagnosis system of fault in automatic measurement system is taken as an analysis case. By comparing the simulated results and actual results of RPN, the method is verified to be the feasible.