Abstract: The detailed spatial distribution of soil properties which is essential for watershed modeling and scenario analysis,is mainly acquired through field soil sampling and digital soil mapping. Especially in the red soil region of southern China where the distribution of soil parent material is complex and with large spatial variability, the acquisition of detailed spatial distribution of soil properties is often one of the main bottlenecks in watershed modeling and scenario analysis, due to both the cost of field soil sampling soil sampling and predictive soil property mapping can effectively reduce the required number of soil samples and predict the detailed spatial distribution of soil properties. To explore the applicability of this method over the red soil region in southern China, we applied this method to conducting purposive soil samplingin a small,red-soil watershed (Zhuxi) in Fujian Province,and then predictive soil property mapping of sand content and organic matter content in the soil of 0 -20 cm at a spatial resolution of 5 m. A set of five topographic attributes ( i. e., elevation, slope gradient, profile curvature, horizontal curvature, and topographic wetness index) were derived from the gridded digital elevation model with 5 m resolution and then were used as environmental variates. Fuzzy clustering method was applied to this set of topographic attributes and got the result of nine fuzzysoil-landscape classes. Purposive soil sampling was carried out at the center of each fuzzy soil-landscape class. Then the value of a soil property at each location can be predicted as the average of the soil property value at every purposive sampling pointweighted by the fuzzy membership value of the location to the fuzzy soil-landscape class represented by the purposive sampling point. The ordinary kriging method with 42 modeling points and a traditional method of linking the typical soil property value to soil-type polygon map were chosen as the comparative methods. Based on the validation with 30 random points independent with the modeling points, the fuzzy clustering method requires only a very few soil samples (only nine modeling points used to build the soil-landscapemodel in this study), and can achieve better prediction accuracy based on the validation with an independent soil sample set. RMSE values of mapping results of sand content and organic matter contentin the soil of 0 - 20 cm are 13.81% and 12.56 g/ kg, respectively. And the predictive soil map from fuzzy clustering method can well reflect the spatial variation of the soil in the study area. Therefore, the fuzzy clustering method is applicable over the red soil region in southern China when the sampling cost of digital soil mapping can be significantly reduced.