Land-surface and vegetation models divide the globe into discrete vegetation classes or plant functional types (PFTs). The current study quantifies some of the limitations of this simplification on global predictions of carbon, water, and energy fluxes. First, a state-of-the-art land-surface model, JULES-SF, is optimized against a diversity of calibration data sets (eddy covariance fluxes, field measurements of net primary production (NPP), and remotely sensed surface albedo) in order to retrieve a range of values for four key plant parameters within each PFT. This is done for 112 sites and 1200 1° land points. Second, global simulations are compared in which the parameter values per PFT are either fixed (standard method) or vary according to either the retrieved parameter range or the satellite-observed range (new methods). Retrieved key plant parameters exhibit a broad range, and the range overlap between PFTs is significant. The impact on the global simulation depends on the surface flux/state in question. Thus, the difference between the new and old method is small for albedo, net shortwave radiation, and continental runoff (0.005, 0.7%, and 2%, respectively) compared to current model-observation differences (0.05, 7%, and 20%, respectively). In contrast, carbon fluxes are more sensitive to the categorization of plant properties, with predicted global NPP varying by ≤15% (6.2 Gt yr−1) according to whether the standard or one of the new methods is implemented.