This paper presents a method to downscale soil moisture at the sub-pixel level. First, we formulated a mixed-pixel model to represent remote sensing (RS) soil moisture in a pixel. Then, we used an un-mixing algorithm to determine the mixing soil moisture signatures contained within the RS pixel. The method involves the use of a local-scale soil-water-atmosphere-plant model (SVAT) to simulate the local scale processes within the remote sensing pixel, and a genetic algorithm to search for the combinations of soils contained within the RS pixel. The simulated local-scale soil moistures are aggregated based on the proposed combinations of soils within the pixel. This aggregated signature in turn is compared with the observed RS soil moisture. Matching these two soil moisture signatures was accomplished by genetic algorithms. Additional by-products of this analysis are the local-scale soil hydraulic parameters for each soil type extracted and the area fractions of soils contained within the pixel. In this paper, we present the development and application of the mixing/un-mixing algorithm, and the numerical and actual case studies used to test the feasibility of the proposed approach to downscale RS soil moisture.