Acquiring data for the soil water content (theta) is important for assessing vegetation growth, drought, and climate change. However, it is a time-consuming and labor-intensive task that is especially challenging in deep soils. Therefore, we introduce the noninvasive technique of electrical resistivity tomography (ERT) for indirectly determining theta through the development of prediction models for loessial regions under different site conditions (i.e., soil texture, land use, soil depth, and dry/wet conditions). We obtained 2769 pairs of electrical resistivity (rho) and theta datasets using ERT (53 sites) and a neutron probe (69 access tubes) on the Chinese Loess Plateau. We built linear and nonlinear models correlating rho and theta and selected the best model according to the coefficient of determination (R-2) and root-mean-square error (RMSE). The uncertainty and sensitivity of ERT-derived theta were further evaluated and acceptable results were obtained. The new models correlating rho and theta under different site conditions are the first set of models based on field data from a loessial region, and their acceptable performance makes them applicable for measuring different soil parameters in the Loess Plateau and possibly other loessial regions around the world.