The grain size characteristics and spatial distribution of topsoil particles could provide references for the monitoring and evaluation of surface wind erosion. The stacked fine gravel Gobi in the northwest of Jartai Salt Lake was selected as the research objective. Based on the digital image processing technique, the percentage contents of different size particles were obtained for the Gobi surface. The spatial heterogeneities of different size particles were analyzed through both classical statistics and geostatistics. The results showed that the digital image processing technique was an effective approach to extract the information of surface particle composition, the total percentage of three size particles of non-erodible, semi-erodible and high-erodible was 93.12%, the mean contents of them were 69.81%, 14.99% and 8.32% with the variation coefficients of 14.12%, 40.02% and 55.06%, respectively. The variation functions of non-erodible and semi-erodible particle contents were classified as Gaussian model, while that of high-erodible particle content as Exponential model. The spatial heterogeneity scale of the three size particles were 632 m, 691 m and 1 875 m, respectively. All the above three indices indicated that the smaller the particle size, the larger the spatial variability. All the three size particles were distributed as patches, and the value of C/(C₀+C) varied from 72.79% to 79.37%, indicated that the spatial pattern of surface particles of Gobi was the combined effect of structural factors and random factors, but the structural factors, such as the forming process of Gobi, terrain, the distribution of wind erosion and vegetation, were the dominant driving forces of the spatial distribution of surface particles.