Mie scattering models suggest that the thermal infrared emissivity of snow is relatively insensitive to variations in its properties and is dependent, primarily, on viewing angle. This gives rise to the possibility of accurately measuring snow surface temperatures, over the polar ice sheets, using satellite infrared radiometers operating in the window region at 10--13 microns. These instruments were designed, primarily, to measure sea surface temperature and a substantial body of work has been undertaken on correcting for the effects of variable absorption by the atmosphere over oceans. The atmospheric conditions over the Antarctic and Greenland ice sheets are significantly different, however, and require special treatment. A three year dataset of radiosonde measurements, collected from six Antartic stations, is used to investigate the behavior of the ''Split-window'' algorithm. The same dataset has been used to test the performance of a dual-view algorithm that can be used with the Along Track Scanning Radiometer onboard ERS-1. It is shown that, given accurate emissivity estimates, the atmospheric correction has an rms error of 0.015 K using the dual-view method. Combined with the excellent calibration and stability of the Along Track Scanning Radiometer and pixel averaging to reduce the detector noise it is possible to derive snow surface ''skin'' temperatures to an accuracy of about 0.1 K. ¿ American Geophysical Union 1994