Micrometeorological measurements of evapotranspiration (ET) can be difficult to interpret and use for validating model calculations in the presence of land cover heterogeneity. Land surface fluxes, soil moisture ($theta$), and surface temperatures (Ts) data were collected by an eddy correlation-based tower located at the Orroli (Sardinia) experimental field (covered by woody vegetation, grass, and bare soil) from April 2003 to July 2004. Two Quickbird high-resolution images (summer 2003 and spring 2004) were acquired for depicting the contrasting land cover components. A procedure is presented for estimating ET in heterogeneous ecosystems as the residual term of the energy balance using Ts observations, a two-dimensional footprint model, and the Quickbird images. Two variations on the procedure are successfully implemented: a proposed two-source random model (2SR), which treats the heat sources of each land cover component separately but computes the bulk heat transfer coefficient as spatially homogeneous, and a common two-source tile model. For 2SR, new relationships between the interfacial transfer coefficient and the roughness Reynolds number are estimated for the two bare soil--woody vegetation and grass--woody vegetation composite surfaces. The ET versus $theta$ relationships for each land cover component were also estimated, showing that that the woody vegetation has a strong tolerance to long droughts, transpiring at rates close to potential for even the driest conditions. Instead, the grass is much less tolerant to $theta$ deficits, and the switch from grass to bare soil following the rainy season had a significant impact on ET.