Scenarios based on 14 different combinations of model structure and parameterization were ranked based on their mean value of Nash–Sutcliffe Efficiency. Results illustrated considerable variability in model performance both within and between biome types. Indeed, no single model consistently outperformed any other when considered across all biomes. For instance, in grassland and shrubland sites, the single-source Penman–Monteith model performed the best. In croplands it was the three-source Mu model, while for evergreen needleleaf and deciduous broadleaf forests, the Shuttleworth–Wallace model rated highest. Interestingly, these top ranked scenarios all shared the simple lookup-table based surface resistance parameterization of Mu et al. (2011), while a more complex Jarvis multiplicative method for surface resistance produced lower ranked simulations. The highly ranked scenarios mostly employed a version of the Thom (1975) formulation for aerodynamic resistance that incorporated dynamic values of roughness parameters. This was true for all cases except over deciduous broadleaf sites, where the simpler aerodynamic resistance approach of Mu et al. (2011) showed improved performance.