Ground-penetrating radar (GPR) can be used to obtain information about the variation in the dielectric constant of the subsurface. The sensitivity of a soil's dielectric constant to the presence of water therefore makes water content estimation by GPR possible. The dielectric constant derived from GPR data, however, is also influenced by the geometric distribution of water in the subsurface. We show that this causes the relationship between the dielectric constant and water content to become scale dependent in complex geologic systems. We have derived dielectric constant-water content relationships that account for subsurface geometries in spatially correlated random media that can be characterized using geostatistics. From these relationships we illustrate that the importance of scale effects are strongly dependent on the variance and the anisotropy of the water content in the subsurface; in some cases, ignoring scale effects will not significantly impact the estimation of water content, while in other cases, large biases can occur. This work provides a conceptual framework for the predictive modeling of field-scale dielectric constant-water content relationships.