Hydraulic pump degradation feature extraction is a key step of condition-based maintenance. In this article, a novel method based on local characteristic-scale decomposition (LCD) and discrete cosine transform–composite spectrum (DCS) fusion algorithm is proposed. In order to reduce noises and other disturbances, vibration signals are first processed by LCD with the high frequency harmonic. Detail components with sensitive information are achieved by the selection of intrinsic scale components. Furthermore, on the basis of the earlier composite spectrum (CS), the DCS fusion algorithm is proposed to make fusion of the obtained detail components for improving the feature performance. The DCS entropy is extracted as the fault degradation feature. Analysis of the hydraulic pump degradation experiment demonstrates that the proposed algorithm is feasible and effective to indicate the performance degradation of the hydraulic pump.