We used thermal imaging in conjunction with the eddy covariance technique to characterize canopy evapotranspiration (ET) from a small heterogeneous grassland. We compared ET estimated by a simple soil–vegetation–atmosphere transfer (SVAT) at field scale (a few 100m²) with that estimated by the eddy covariance method. These two independent estimates of ET showed a good correlation when the flux source area was the same. However, whereas the eddy covariance method yielded integrated results over a large, variable landscape area, the SVAT model primarily yielded values reflecting just the grassland area. We estimated mapped transpiration (Tr) at a point scale (1m²) and showed that Tr increased linearly with increasing leaf area index (LAI). Although stomatal conductance of C₃ plants was appreciably larger than that of C₄ plants at the leaf scale, this difference was not reflected in Tr at the canopy scale. Tr may be more sensitive to aerodynamic conditions (wind speed and radiation) or environmental heterogeneity (soil–water) than stomatal conductance. The SVAT model clarified variations in the spatial distribution of Tr over a heterogeneous grassland.