Abstract: The fault features are difficult to be extracted from the worn rolling ball bearings. To tackle this problem, an algorithm of the Volterra series kernel based on the multiple-pulse excitation method was proposed. This method belongs to the cross diagnosis for nonlinear system models, which utilizes the sampled signal input and output of the bearing system to establish the Volterra nonlinear identification system model and applies the Volterra low-order kernel algorithm based on the multiple-pulse excitation method to obtain the low-order kernel, then the low-order kernel will be compared in aspects of the GIRF and the GFRF to estimate the present running state of the bearing system. The major bearing of a centerless lathe was taken for example to verify this method through experiment. In contrast to the traditional wavelet analysis method, the multiple-pulse excitation method helps extract the fault features of the ball bearing conveniently and exactly. Thus, the proposed method has much significance to the diagnosis of such faults.