Different materials have different absorption characteristics, so their spectral reflectances have different characteristics. In order to analyze the pigments of Dunhuang murals, an algorithm was designed to recognize pigments automatically based on extracted features of their visible reflectance spectra. The first and second order derivative of the visible reflectance spectrum in the range 400~700nm was computed to extract peak information. The absolute value of the negative position in second order derivative was selected as feature data to present the peaks. A median pyramid was used to derive three resolution representations of the featured curves to enhance the most important information, and then these three curves were normalized and combined as the feature space F.The Euclidean distance between sample points and pigments database was calculated as the similarity to identify the mineral pigment of target color points. Results show that the nondestructive identification method can recognize the pigments of Dunhuang murals efficiently and accurately. The method is worth applying to the restoration and digital archiving of murals.