Prediction of large-scale vadose zone solute transport is affected by errors due to uncertainties in model structure (model complexity) and model input (parameter values, boundary and initial conditions). Selection of an appropriate level of model complexity must consider these two sources of uncertainty. This paper illustrates this selection process for the prediction of root-zone average salinity at different spatial scales, ranging from point to field scale. Salt transport is simulated with two models of different complexity. The first “complex” model is based on the numerical solution of the one-dimensional Richards and advection-dispersion equations. The second “simple” model consists of newly derived analytical solutions to the depth-integrated flow and salt balance equations, assuming a time-invariant relationship between salt storage and drainage load. Statistical moments of root-zone average salinity predicted with the “simple” and “complex” models are compared for different levels of model input uncertainty, ranging from point scale (small uncertainty) to field scale (large uncertainty). Results show that the “complex” model may be replaced by the “simple” model as long as model input uncertainty is large enough.