In order to extract the fault features of rolling bearing effectively, a new improved enhanced kurtogram method is proposed. Improved enhanced kurtogram is calculated based on harmonic wavelet packet composition and the node whose kurtosis value is maximum is selected after calculating the improved enhanced kurtogram of the original fault signal, then reconstruct the signal through the harmonic wavelet packet coefficient of the optimal node, the rolling bearing fault type could be judged by analyzing the envelope spectrum of the reconstructed signal. The comparison of the proposed method with the original kurtogram method and the enhanced kurtogram method are conducted to analyze the experimental signal of rolling bearing. The results show that the new method proposed in this paper could select the resonance frequency band precisely and could be applied effectively on fault diagnosis for rolling bearing.