Abstract: A new spectral correlation density method based on kurtosis energy is proposed in this article. In the method the kurtosis of every slice along the cyclic frequency axis is calculated and used as the weight coefficient to evaluate the modulation ability of the corresponding cyclic frequency, and then the fault feature is effectively extracted by energy accumulation of the cyclic frequency. Compared with the traditional spectral correlation density (CSD) method, the proposed method has great performance of reduction the influence of multiple-frequency harmonics on the characteristic fault frequency, so the fault feature is extracted more clearly and accurately. Simulation signals and the fault signals from the outer ring of a rolling bearing in a low-speed and heavy-duty test bed have been analyzed by CSD, the proposed method and resonance demodulation. The analysis results prove that the proposed method is effective.