A procedure for estimating sea surface temperature was developed by using the near and thermal infrared data available from full resolution (1.1 km field of view) daytime passes of the NOAA 6 satellite Advanced Very High Resolution Radiometer. It is demonstrated that spatially unresolved clouds and variable amounts of water vapor in the atmosphere are the major error sources. The procedure therefore incorporates a means of identifying radiometer fields of view which are virtually cloud-free, by comparing measured solar radiation at 1 μm, backscattered from the atmosphere, against that expected from a simple scattering model. A set of 82 cloud-free fields of view is then assembled, spatially and temporally nearly coincident with research quality (¿0.2¿C) in situ ship and buoy sea surface temperature reports collected during 1979 and 1980. The set is then used to develop and evaluate a simple model to reduce variable water vapor effects by relating in situ sea surface temperature to the infrared brightness temperatures measured by the radiometer at 3.7 and 11 μm. Residual disagreement between satellite estimates and in situ measurements of sea surface temperature is about 0.6¿C for the data set which extends from the equator to 50¿N in the northeast Pacific. The procedure is then applied to 10 satellite passes in the northeast Pacific during April 1981 to obtain monthly mean sea surface temperature and temperature anomaly (departure from climatological norm) maps. These maps are compared against similar ones produced from routine ship reports, mostly consisting of engine injection temperature data, typically good to no better than 1¿C. The maps agree to ¿0.8¿C (1 sigma), with the level of agreement clearly limited by the poor accuracy of the ship reports. It now appears that the radiometer data, when properly handled, can be used as a more accurate substitute for routine ship reports of sea surface temperature in applications such as climate variability studies requiring 0.5--1.0¿C accuracy.