Soil moisture distribution shows highly variation both spatially and temporally. This study assesses the spatial heterogeneity of soil moisture on a hill-slope scale in the Loess Plateau in West China by using a geostatistical approach. Soil moisture was measured by time-domain reflectometry(TDR) in 313 samples. Two kinds of sampling scales were used(2 x 2 m and 20 x 20 m) at two soil layers(0-30 cm and 30-60 cm). The general characteristics of soil moisture were analyzed by a classical statistics method,and the spatial heterogeneity of soil moisture was analyzed using a geostatistical approach. The results showed that the spherical model is the best-fit model to simulate soil moisture on the experimental hill-slope. The parameters of this model indicated that the spatial dependence of soil moisture in the selected hill-slope was moderate. Even the 2 x 2 m sampling scale was too coarse to show the detailed spatial variances of soil moisture in this area. The dependent distance increased from 27.4 m to 494.16 m as the sampling scale became coarse(from 2 x 2 m to 20 x 20 m). A map of soil moisture was generated by using original soil moisture data and interpolated values determined by the Kriging method. The average soil moisture(area weighted) in the different layers of soil was calculated on the basis of this map(10.94% for the 0-30 cm soil layer,11.88% for the 30-60 cm soil layer). This average soil moisture is lower than the corresponding average effective soil moisture,which suggests that the soil moisture is not sufficient to support vegetation in this area. Soil moisture distribution shows highly variation both spatially and temporally. This study assesses the spatial heterogeneity of soil moisture on a hill-slope scale in the Loess Plateau in West China by using a geostatistical approach. Soil moisture was measured by time-domain reflectometry (TDR ) in 313 samples. Two kinds of sampling scales were used (2 x 2 m and 20 x 20 m) at two soil layers (0-30 cm and 30-60 cm). The general characteristics of soil moisture were analyzed by a classical statistics method, and the spatial heterogeneity of soil moisture was analyzed using a geostatistical approach. The results showed that the spherical model is the best-fit model to simulate soil moisture on the experimental hill-slope. dependence of soil moisture in the selected hill-slope was moderate. Even the 2 x 2 m sampling scale was too coarse to show the detailed spatial variances of soil moisture in this area. The dependent distance increase d from 27.4 m to 494.16 m as the sampling scale became coarse (from 2 x 2 m to 20 x 20 m). A map of soil moisture was generated by using original soil moisture data and interpolated values ​​determined by the Kriging method. The average soil moisture (area weighted) in the different layers of soil was calculated on the basis of this map (10.94% for the 0-30 cm soil layer, 11.88% for the 30-60 cm soil layer). This average soil moisture is lower than the corresponding average effective soil moisture, which suggests that the soil moisture is not sufficient to support vegetation in this area.