Accurate, spatially distributed surface temperatures are required for modeling evapotranspiration (ET) over agricultural fields under wide ranging conditions, including stressed and unstressed vegetation. Modeling approaches that use surface temperature observations, however, have the burden of estimating surface emissivities. Emissivity estimation, the subject of much recent research, is facilitated by observations in multiple thermal infrared bands. But it is nevertheless a difficult task. Using observations from multiband thermal sensors, ASTER and MASTER, estimated surface emissivities and temperatures are retrieved in two different ways: the temperature emissivity separation approach (TES), and the normalized emissivity approach (NEM). Both rely upon empirical relationships, but the assumed relationships are different. TES relies upon a relationship between the minimum spectral emissivity and the range of observed emissivities. NEM relies upon an assumption that at least one thermal band has a predetermined emissivity (close to 1.0). Experiments comparing TES and NEM were performed using simulated observations from spectral library data, and with actual data from two different landscapes-- one in central Oklahoma, USA, and another in southern New Mexico, USA. The simulation results suggest that TES's empirical relationship is more realistic than NEM's assumed maximum emissivity, and therefore TES temperature estimates are more accurate than NEM estimates. But when using remote sensing data, TES estimates of maximum emissivities are lower than expected, thus causing overestimated temperatures. Work in progress will determine the significance of this overestimation by comparing ground level measurements against the remote sensing observations.