AbstractTropical vegetation is a major source of global land surface evapotranspiration, and can thus play a major role in global hydrological cycles and global atmospheric circulation. Accurate prediction of tropical evapotranspiration is critical to our understanding of these processes under changing climate. We examined the controls on evapotranspiration in tropical vegetation at 21 pan‐tropical eddy covariance sites, conducted a comprehensive and systematic evaluation of 13 evapotranspiration models at these sites, and assessed the ability to scale up model estimates of evapotranspiration for the test region of Amazonia. Net radiation was the strongest determinant of evapotranspiration (mean evaporative fraction was 0.72) and explained 87% of the variance in monthly evapotranspiration across the sites. Vapor pressure deficit was the strongest residual predictor (14%), followed by normalized difference vegetation index (9%), precipitation (6%) and wind speed (4%). The radiation‐based evapotranspiration models performed best overall for three reasons: (1) the vegetation was largely decoupled from atmospheric turbulent transfer (calculated from Ω decoupling factor), especially at the wetter sites; (2) the resistance‐based models were hindered by difficulty in consistently characterizing canopy (and stomatal) resistance in the highly diverse vegetation; (3) the temperature‐based models inadequately captured the variability in tropical evapotranspiration. We evaluated the potential to predict regional evapotranspiration for one test region: Amazonia. We estimated an Amazonia‐wide evapotranspiration of 1370 mm yr−1, but this value is dependent on assumptions about energy balance closure for the tropical eddy covariance sites; a lower value (1096 mm yr−1) is considered in discussion on the use of flux data to validate and interpolate models.