During the POLYMODE experiment, infrared and visible data from GOES were the only satellite data permanently archieved in digital form. GOES data lack onboard calibration capability and were fitted, using least squares and other methods, to XBT, buoy tide gage, and ship of opportunity reports to remove atmospheric effects and calibration offset. For the experiment period considered, December 1977 through February 1978, bivariate Gaussian discriminant function cloud identification showed more than 93% of the 8-km resolution GEOS infrared pixels to be cloudcontaminated. Cloud-free in-situ calibration points were distributed in nonrandom groups, which resulted in systematic errors when using least squares techniques. Surfaces and regression lines were least squares fitted between satellite and in-situ data: differences and ratios were also used. Best results were obtained with a regression in the form of the infrared radiative transfer equation but were no better than ¿0.9 K. As a result of extensive cloudiness the linear regressions were seldom useful, and temperature ratios with ¿1.3 K experimental errors best represent the applicability of GOES data to sea surface temperatures. A composite of all cloud-free data for the three months left 5% of the area unobserved, but the mean surface temperature map and the standard deviation map are an improvement over atlas maps because of higher spatial resolution.