The utility of an unsaturated zone soil moisture model is not only its ability to describe the soil moisture dynamics at a given point but also the possibility to generalize the results to larger areas. In this study we investigated the predictive performance of a one‐dimensional unsaturated zone soil moisture model when applied at point, field, response unit, and catchment scales, using detailed field observations from a 0.42‐km2 catchment in the Netherlands. Our main question was how model parameterization and model performance could be compared across these scales. We considered two different calibration–validation schemes and three performance statistics. In all cases we applied the same Levenberg–Marquardt optimization scheme. Differences between calibration–validation schemes (interpolation vs. extrapolation) were surprisingly small. Using one particular model parameterization across the various aggregation levels, the optimal Mualem–van Genuchten parameters for a coarser aggregation level can be derived from an underlying level by simple arithmetic averaging. The different performance indices (RMSE, index of agreement, and Nash–Sutcliffe coefficient) were highly variable between observation locations and for different aggregation levels. Overall, the indices were more favorable at higher aggregation levels, and in correspondence with errors reported in comparable studies. The unsaturated‐zone model did not, however, provide satisfactory predictions of independent flux observations, in this case daily catchment discharge. Moreover we did not succeed in deriving a meta‐model to scale model performance indices with aggregation level. Our case study therefore supports the view that multiscale calibration studies that use both state and flux observations are required to compare results from unsaturated zone models at different aggregation levels.