The relationship between topography, land use, and topsoil moisture storage is investigated for a small catchment with undulating deep loess hilslopes in the south of the Netherlands. For a period of 10 months, soil moisture profiles have been measured weekly at 15 locations throughout the catchment. A Generalized Additive Model was employed to find relationships between the various factors influencing soil moisture. It defines a water balance as a sum of non-linear components. The water balance is applied to our data at various spatial (catchment, response unit, hillslope and plot), and temporal (monthly, weekly and daily) scales. Each of the water balance components is parameterized as a function of topographic variables, land use variables, weather variables and antecedent soil moisture. The model framework is hierarchical: it starts at the coarsest spatio-temporal resolution, the water balance components found here act as constraints when identifying models at finer resolutions. It turns out that the importance of land-use variables varies considerably with temporal resolution. At coarse resolutions land-use is unimportant, whereas at finer resolutions it becomes more relevant. Land use is equally important over all spatial resolutions (response unit and finer). Topography is mostly relevant at the plot scale. The water balance terms become increasingly non-linear at finer scales. Evapotranspiration depends mainly on reference evapotranspiration and crop cover. Drainage to deeper layers depends mainly on soil moisture and to a lesser extent on topography. Lateral transport is weakly dependent on topography. It appears that autoregressive components become increasingly important at finer temporal resolutions.