The measured vibration signal is polluted easily by the noise so that useful information is drowned. When one-dimensional signal filtered by mathematical morphology filter, the flat structure element length is selected in experience and randomness way. Concerning above problems, a new method based on morphological difference operator, power spectral entropy theory and feature energy ratio is proposed. The method is used to determine the best length of flat structure element so as to achieve the optimal filtering effect. Firstly, one-dimensional signal is filtered by morphological difference operators with different flat structure element lengths. Secondly, the power spectral entropies and feature energy ratios of the signals filtered are computed. Lastly, according to power spectral entropy and feature energy ratio theory, the result is that the length corresponding to a minimum of the power spectral entropies is the optimal one. The simulation signal and fault signal of the hydraulic pump are tested and verified, morphological difference operator with the optimal length range of flat structure element can filter fault signal of hydraulic pump best and retain more fault characteristic information.