In this study we aim to (1) explore the differences in the accuracy of satellite-derived land-surface skin temperature for day and nighttime observations, (2) assess the effects of large solar zenith angles, and (3) develop an understanding of the spatial variability of the observed temperatures. Land-surface skin temperatures are obtained using the split-window technique from observations of the AVHRR instrument aboard the NOAA-12 and NOAA-14 satellites for the SGP97 (Southern Great Plains 1997) hydrology experiment. From the study of several days of observations we find that observed biases with respect to the ground temperature, both during day and night, are small. However, except for one rainy day measurement, there consistently was a warm bias during the day and cold bias during the night. Contrary to the hypothesis that at large solar zenith angles the observed temperatures are representative of the shelter height air temperature, we find that the observed temperatures are still closer to the ground temperatures than the air temperature. The spatial correlation is nearly isotropic and has an exponential decay. The correlation lengths demonstrate tremendous spread both during the day and during the night with the mean during the day being about 50% larger than that during the night. The average correlation length of 8.63 km during the day is much smaller than the grid sizes typically used in short-time hydrology/mesoscale forecasts. This suggests that for modeling purposes the temperatures in each grid box may be treated as uncorrelated. However, the variance captured can be significantly smaller than the true value. ¿ 2000 American Geophysical Union