A method of the fault detection and diagnosis by combining wavelet correlation filter and Shannon information entropy was proposed. The method was called “wavelet correlation feature scale entropy-fault”. At first, the weak fault information features were picked up from roller bearings fault vibration signals by the way of wavelet transform correlation filter, in order to get high signal to noise scales wavelet coefficients. Then, the defining and computing way of wavelet correlation feature scale entropy was presented, based on the integration of shannon information entropy theory. Wavelet correlation feature scale entropy can quantitatively describe energy distributing of different scales which reflects difference of roller bearings running state, the work states and fault types were estimated by magnitude of the selected wavelet correlation feature scale entropy which can mostly embody fault features. Results and analysis of the diagnosing example reveal that the proposed method can effectively estimate roller bearings fault feature and offer one new way for diagnosing roller bearings.