Abstract: In order to solve the fault diagnosis problem of wind turbine rolling bearings,a novel intelligent fault diagnosis method based on imprvoed multi-class relevance vector machine（multiclass relevance vector machine,MRVM）was proposed.Firstly,a new technique to optimize MRVM kernel function parameters was adopted to eliminate the subjectivity of parameter selections,and then the simulation experiment results verify the effectiveness and superiority of this technique.Finally,the spindle bearings fault diagnosis experiment was implemented. The wavelet packet energies of rolling bearing vibration signal were extracted as fault features. Experimental analysis show that the proposed method not only can improve the accuracy and efficiency of fault diagnosis,but also can output the probability information of fault diagnosis,which the probability information provide a reference for maintenance staff. In addition,the comparison experiments with other methods further indicate the superiority of the proposed method in intelligent fault diagnosis.