The study on spectral mixing mechanism has a certain instructive significance to unmixing. With a full-wavelength spectrometer, the research made a controlled acquisition of the spectral reflectivity of pure snow pixels, pure desert-vegetation pixels and snow-desert vegetation mixed pixels in the mode of rule and irregular distribution during the accumulation period and ablation period. The ratio of snow area to desert vegetation area of images was calculated by K-means clustering algorithm and spectral variation characteristics of mixed pixels were analyzed; to obtain more precise spectral characteristic information, the absorption characteristic parameters to the response bands were calculated; the images were collected by quasi-synchronous Tetracam ADC3 and the typical indices were calculated. It’s verified at the micro-scale that the mixed pixels are mainly located at the boundary between one category to another. The results are seen as follows: the spectral reflectivity of coarse-grained frozen snow is obviously higher than that of new snow which is obviously higher than that of aged snow in the ranges of 1 456～1 697 nm. Because of the ice cover, the collected desert vegetation spectra is actually a mixture of spectral information of snow, ice crystals and vegetation branches; the spectral properties of vegetation covered with new snow are actually the mixed spectral information of snow and vegetation branches; there is no “red edge” effect like conventional green vegetation. When the acquisition angles are 5° and 10°, the spectral reflectivity is lower than that at a vertical angle. When the acquisition angle is bigger than 10°, the spectral reflectivity increases if the angle becomes bigger.