Considering that the minimum entropy deconvolution (MED) was only sensitive to a single abnormal vibration signal, and the length of the filter needed to be adjusted manually, a maximum kurtosis entropy deconvolution (MKSED) method was proposed and applied to bearing fault diagnosis. Considering that kurtosis entropy has the advantage of continuous shock oscillation, kurtosis entropy was chosen as the objective function of deconvolution. At the same time, kurtosis entropy was used as the fitness function of the improved local particle swarm optimization algorithm (LPSO), and LPSO was used to optimize the filter length, so that MKSED can adaptively adjust the filter length when deconvolution, so as to accurately extract the continuous pulse signal. The experimental results show that the method can extract continuous pulse signal more effectively and improve the accuracy of fault diagnosis.