Early fault characteristics of wind turbine generator is not obvious, less fault data can be collected, and the diagnosis accuracy is relatively low. A method of EEMD combined with RVM was proposed for early diagnosis by multiple faults of wind turbines. Firstly, pre-process the vibration signals of every faults of the wind turbine by EEMD which combined with the grey relational degree, and extract the optimal fault characteristics; then, train the fault characteristics by the RVM for establish the fault diagnosis model. Applying this method to a real diagnosis of wind turbine vibrating fault, by comparing the diagnostic results which obtained by WPD with RVM and EEMD with LS-SVM, the results show that the EEMD-RVM method is feasible and has the advantages of short time consuming and high precision.