Based on the soil survey data of Yulong county，Yunnan province in 2009，the soil organic matter content of the samples was converted to organic carbon density based on the soil type method，spatialized by Kriging interpolation，and then 2009 Landsat7-Level2 image and Shuttle Radar Topographic Mission 90 m DEM data were used to extract the explanatory variables which are closely related to soil organic carbon formation，such as NDVI，brightness，greenness，humidity，slope，aspect，and curvature.The density and spatial distribution of soil organic carbon were simulated through a random forest model，and the total organic carbon storage of 0～20 cm surface soil was estimated based on organic carbon density.The errors of the results by the two methods were analyzed.The results showed that the estimated results of Kriging and random forest were 2.4×108 and 1.7×108 t， respectively and the root-mean-square errors were 20.77 and 14.11，respectively.The ordinary Kriging interpolation error was larger and strongly depended on the number and spatial distribution of sampling points.The random forest model can not only handle high-dimensional data，but also calculate the importance of multiple variables.The accuracy of the estimation result by the random forest model is higher and closer to the actual situation of the region，and the performance of small-scale details is better and suitable.It is used to estimate the density and storage of soil organic carbon in counties with complex terrain and limited sample points.