The shortage of fault characteristic parameters as well as the limitation of fault information in power transformer fault diagnosis may result in an unsatisfied diagnosis result. To cope with such a problem

34 characteristics obtained by the combination with electrical test data and dissolved gas analyses（DGA） were taken as fault parameters to refine the fault characteristics. On such basis

the feature evaluation and kernel principal component analysis（KPCA） based fault diagnosis method was developed by a combination of the two approaches. By sensitive evaluation firstly

insensitive parameters were eliminated and thus help to weaken their influences on characteristics. Then

27 types of characteristics were conducted with the KPCA to reduce their dimensions. Finally

the extracted fault parameters with 9 dimensions were taken as the input vector of multiclass relevance vector machine（M-RVM） for fault classification. Case analysis shows that this method not only can compensate the deficiencies like shortage in fault feature parameters effectively

but also is more general

and fault identification accuracy is increased to 90.35%

which can provide a reference for the fault diagnosis of transformer fault case information limited.