To extract the dynamic response characteristics of blades from the vibr ation response of whole turbine set efficiently, then take it as the characteristic parameters of blade working states and achieve fault detection and diagnosis of the wind turbine blades, a finite element method is used to build a dynamic model for 300 W wind turbine. Natural frequency and vibration mode of whole set can be obtained and compared to validate the rationality of the model by modal anal ysis. Vibration acceleration response signal under normal working state and ecce ntricity fault state can be detected by the hammering method., The length fractal dimension which is taken as the index of diagnosing the eccentricity fault of blades can be calculated using the theory of length fractal dimension in nonline ar vibration system. The result shows that the method is effective to distinguis h the normal state and the eccentricity fault state, and detect the eccentricity posi tion and the value of wind turbine blades.