The current tendency in physically based soil-vegetation-atmosphere transfer (SVAT) schemes is to use increasingly complex process descriptions to predict evaporative fluxes at both patch and landscape scales. This approach does not take proper account of the heterogeneities that are evident in any landscape. A top-down approach to sub-grid-scale land surface parameterization would suggest that the current complexity may not be supported by the calibration data available. By comparing three SVAT schemes of differing complexity within the generalized likelihood uncertainty estimation framework, we demonstrate the utility of simpler rather than more complex models when calibrated against flux data from various intensive field campaigns. A more robust calibration is achieved for a simple evaporative fraction approach allowing the feasible parameter ranges to be more strongly conditioned by the available data. It is argued that a top-down (dominant mode) predictive model based on a database of such robustly estimated parameter values would result in no greater uncertainty at the scales of application than trying to form parameter sets for complex models from a variety of sources.