This study evaluates the skin temperature (ST) datasets among the ISCCP-D satellite product, the ISCCP-FD satellite product, the ERA40 reanalysis, the NCEP-NCAR reanalysis and the NCEP-DOE AMIP-II reanalysis. Each dataset has its pros and cons. All the datasets are correlated to each other with correlation coefficients > 0.50 for monthly anomalies. To evaluate how good their qualities are, the 5 ST datasets are used to calculate clear-sky (CS) outgoing longwave radiations (OLR) and the results are compared with the ERBE satellite observation. It is found that in general, in northern-hemisphere high latitude regions (tundra, wetlands, deciduous needleleaf forests and sea ice), the simulated CS OLRs derived from using the satellite STs have higher correlations and lower root-mean-squared errors in anomalies than those derived from using the reanalyses' STs. ERA40 underestimates the amplitude of the seasonal ST on ice sheets. All the reanalysis products (ERA40, NCEP-NCAR, NCEP-DOE AMIP-II) overestimate the ST on ocean grids in the 40-70°N latitude zone when the grids are iced. Nonetheless, suspected spurious noises with amplitude at 2 K in the satellite-derived STs produce unphysical anomaly to land types of which the amplitude of the anomaly is weak (such as open water bodies, croplands, rainforest, grasslands, hot deserts, cold deserts). Besides, the satellite derived STs tend to bias high on hot deserts, and the reanalyses derived STs tend to bias low on mountainous areas.