Satellite-derived multichannel sea surface temperature (MCSST) data are evaluated for the period 1982--1988 relative to n situ data from the comprehensive ocean-atmospheric data set (COADS). A number of statistical methods are used to assess the quality of the MCSST data. Temporal anomaly cross correlations are significant at the 95% confidence level for all basins and subregions examined, except for the North Atlantic Ocean. Biases in the MCSST product, caused by volcanic aerosols and by use of the wrong satellite calibration tables, caused problems in the utility of the MCSST data for climate monitoring. Signal-to-noise variance ratios range from 0.49 in the North Atlantic Ocean to above 4 in the equatorial Pacific. Mean differences between MCSST and COADS show that the MCSST values are lower than COADS by 0.19¿ to 0.64¿C. The mean differences are found to have an annual cycle that varies with latitude and season. The mean MCSST standard deviations are always lower than those of COADS (except again for the North Atlantic Ocean), ranging from 0.03¿ to 0.45¿C. Accurate monitoring of SST variability in the southern oceans is not possible without the use of the satellite data. Because of the stringent accuracy requirements of SST data for climate monitoring, however, the utility of MCSST data alone for this purpose is only marginal.