The objective of this study was to evaluate the sensitivity of instantaneous latent heat flux (LE) estimation from Surface Energy Balance Algorithm for Land (SEBAL) by changing hot/cold pixel selections. The SEBAL model was programed in a Matlab environment and applied to Lower Fox Watershed in northeast Illinois using two Landsat 5 Thematic Mapper images acquired in summer 2006. Unlike most previous studies where hot/cold pixel were manually selected by image analysts, we emphasized an automated hot/cold pixel selection based on land cover map, normalized difference vegetation index (NDVI) map, and land surface temperature (LST) map. Various combinations of hot/cold pixels were automatically selected along the LST gradient. The LE estimations were then validated against ground-based eddy covariance observation. Results show that the LE estimations from SEBAL were sensitive to both hot and cold pixel selections and tend to be more sensitive to cold pixel selection. The absolute percentage difference (APD) of LE estimation compared with field observation data can range from 0.67% to 67.2% by varying hot and cold pixel combinations. The location of hot/cold pixels appears to have minor impact on SEBAL LE estimation. The LE estimation become acceptable (APD < 10%) when using the hot/cold pixels with a slightly higher/lower LST than LST extremes from the study area. This study provides insights into hot/cold pixel selection and the sensitivity of SEBAL-based LE estimation. Future research on SEBAL ET estimation should focus on enhancing automated hot/cold pixel selection algorithm to improve the model's operational use. / Master of Science