Current 'physically based' soil-vegetation-atmosphere transfer (SVAT) schemes use increasingly complex descriptions of the physical mechanisms governing evapotranspiration fluxes, thereby requiring the specification of a large number of parameters controlling the vertical fluxes over a single homogeneous area. Recent attention towards the incorporation of sub-grid scale spatial variability in SVAT parameterisations promises to increase the number of parameters for these models. In this paper, it is demonstrated that a simple patch scale SVAT model still permits too many degrees of freedom in terms of fitting the model predictions to calibration or validation data; it is shown that good model fits may be achieved in many areas of the parameter space. Using a Monte Carlo framework, a sensitivity analysis is performed for simulations of data sets from FIFE and Amazonian sites. This is employed to evaluate the role of each parameter for each forcing dataset, and to identify the controlling and redundant parameters and processes. The results suggest that equifinality of parameter sets in calibration to field data must be expected, that there will be a consequent uncertainty in predictive capability and that more emphasis will be required on identifying the critical controls on evapotranspiration in extending predictions from patch to landscape scale in different environments.