Spatial variations in sand composition were mapped on a regional scale in a terrestrial sand sea, the Gran Desierto of Sonora, Mexico. Mesoscale mapping on a satellite image base allowed quantitative interpretation of the dynamic development of sand sheets and dunes. The results were used to interpret the Quaternary geologic history of the tectonically active region at the mouth of the Colorado River. Landsat thematic mapper multispectral images were used to predict the abundance of different mineralogies of sand grains in a mixed aeolian terrain. A spectral mixing model separated the effects of vegetation and topographically induced shading and shadow from the effects produced by different mineral and rock types. Compositions determined remotely agreed well with samples from selected areas within the spectral limitations of the thematic mapper. A simple discrimination capability for active versus inactive sand surfaces is demonstrated based upon differences in the percentage of low-albedo accessory grains occurring on dormant aeolian surfaces. A technique for discriminating between low-albedo materials and macroscopic shade is implemented by combining thermal images with the results of the spectral mixing model. The image analysis revealed important compositional variations over large areas that were not readily apparent in the field.