Spatial heterogeneity of soil water retention curves (SWRC) plays an important role in modelling the movement of water and solutes in soils. To characterize the spatial variability of SWRC parameters fitted by the van Genuchten (VG) equation, and then identify factors that closely correlated with the VG parameters, we collected undisturbed and disturbed soil samples from the upper 0–5 cm soil layer at 382 sampling sites across the entire Loess Plateau of China ( 620 000 km2). We determined the SWRC and six other soil properties as well as nine environmental factors for each site. Results showed that the saturated water content (us) and a curve-shape parameter that is related to soil pore size distribution (n) were weakly spatially variable (coefficient of variation (CV) ¼ 15%), and that the residual soil water content (ur) and a scaling parameter that is related to the inverse of the air entry pressure (a) were moderately (CV ¼ 96%) and strongly variable (CV ¼ 136%), respectively. Semivariograms of ffiffiffiffi pur and log(a) were best-fitted by an isotropic exponential model, while those of us and log(n) were fitted by an isotropic spherical model. Fractal analysis indicated that log(a) demonstrated the most short-range variation, while log(n) reflected the importance of long-range variation. Bulk density and the contents of clay, silt, sand and soil organic carbon contributed greatly to the variations in the VG parameters (except a). However, the vegetation, topographic and climatic elements as well as land use also usually explained some of the variation. The VG parameters, at the regional scale, demonstrated different characteristics of spatial variation, which comprehensively reflected the combined effects of processes involving the soil, vegetation, topography and climate as well as effects of land use, which is often determined by human activity.