Abstract: Based on the spectral analysis of snow, soil, vegetation and cloud, it is pointed out that the first two channels of traditional NOAA-AVHRR have troubles to distinguish snow from cloud, and the low reflectance of snow in 1.6 μm infrared channel can be used not only to distinguish snow from cloud but also to supply more spectral information to extract snow cover. So the principal components analysis (PCA) was made to AVHRR data, and it was found that the PCA-transformed first two principal components (PCA1, PCA2) contributes about 99% cumulative variance. And the scatter plot to these components was analyzed and the endmembers were given. Also, two different methods were adopted to extract snow cover by using spectral linear mixing model, and results are highly consistent, which indicate that the unmixing method is an effective way to retrieve snow cover parameter.