Based on geo-statistical principle and semi-variability theory, this paper quantificationally studied the spatial variability of soil factor for modeling of soil erosion by water in GIS, comparing the accuracy of different interpolation methods, including IDW，Spline and Ordinary Kriging. The results showed that the spatial correlation of regional soil coefficient of soil anti-scourability was moderately, the nugget-to-sill ratio was 32. 29%，while the spatial correlation of steady infiltration rate，collapse velocity and shear strength was very strongly correlated，the nugget-to-sill ratio was 13. 19%，11. 61 % and 12. 98%respectively. In a word，taking account of MRE, RMSE and interpolation results, Ordinary Kriging interpolation was the best one,which represented the spatial characteristics of soil factors well and can be better used in regional soil erosion model in view of data requirement. To Ordinary Kriging, the lag step of the steady infiltration rate was 30000 meters, and the semi-variance theory model was exponential model whereas the soil coefficient of restitution,collapse velocity and shear strength was 50 000 meters respectively, and the semivariance theory was Gaussian Model. Besides, owing to the spatial distribution of soil type ranging from north to south, each index showed clear zonation differentiation character.