Taking Xitiaoxi basin, Zhejiang Province as the study area, a digital soil mapping technique was introduced in this paper based on artificial neural netweork method. The study integrated comprehensively regional topography and soil types and collected 43 soil profiles for model development and validation. Using the radial basis function (RBF) neural network, the study established a nonlinear relation between six terrain factors, i.e., elevation, slope, plan curvature, profile curvature, surface slops intensity, composite terrain index and soil particle composition and then simulated the space distribution of soil particle composition. Validation results showed that the RBF neural network method could provide a reliable spatial prediction for soil mapping. The model performance of RBF neural network as shown by prediction accuracy and stability is acceptable, which shows that the method is a mapping tool of low cost and high efficiency in the hilly area.