Application of GOES archived data to determing sea surface temperature in the Gulf of Mexico is investigated for October 1977, a period when two research ships and three enviromental buoys were available to provide surface calibration data. A theoretical error analysis is first useed to explain the persistent differences between ship and operationally derived satellite temperatures and then is used to dictate a technique for applying full resolution archived GOES visible and infrared data to the problem. The technique developed first remaps the satellite scanline data into a rectangular matrix that covers the region of interest. Second, radiative transfer calculations are performed at each ''cloud-free'' station around the Gulf's periphery using LOWTRAN-4. Third, least square polynomial surfaces of atmospheric corrections are fitted to the perimetric radiative transfer results. Fourth, a bivariate Bayesian discriminant function, which uses cloud-free areas as training sets, is used to classify objectively clear ocean radiance measurements. Fifth, the GOES is calibrated by comparing in situ measurements with calculated theoretical satellite temperatures. Sixth, maps of sea surface temperature are produced and compared with ship reports and a marine atlas. Root-mean-square calibration errors of GOES surface temperature determined by the above outlined approach was ¿1.2 K when using all available surface reports and ¿0.5 K when using only research ship, XBT, an enviromental buoy reports. Evaluation of infrared multispectral approach planned for NOAA-C shows that for a mesoscale area the two techniques result in similar errors, and that ¿1.0 K or slightly less is probably the geophysical limit of accuracy.