Two independent daily stratospheric data sets are compared for 16 northern winters. The objective is to assess the consistency of temperatures low enough for polar stratospheric cloud (PSC) formation at 50 hPa. The first data set is the subjective analysis produced from the radiosonde network at the Freie Universit¿t Berlin (FUB), which is constrained by hydrostatic and thermal wind balance. The second is the satellite-based analysis of geopotential height, compiled from the TIROS Operational Vertical Sounding system by the United Kingdom Meteorological Office; temperatures are derived from the hypsometric equation. The stratospheric sounding units (SSU) provide most of the stratospheric data in that system. The FUB data are generally colder, particularly at low temperatures, but there is a large dispersion about the mean difference. The uncertainties of the values of the lowest temperatures are around 1 K and 2 K in the mean and rms, respectively. There may be a geographical bias in the data sets. There is a clear relationship between the vertical temperature gradient and the difference between the two data sets, the satellite-derived values becoming relatively colder when the temperature decreases at pressures lower than 50 hPa. Regarding PSC formation: adequately low temperatures occur more often in the FUB data, but on 25% of winter days the areas A&tgr; where PSCs might form are larger in the SSU data. Seasonally integrated values of A&tgr; show a fairly good agreement between the two data sets, the satellite-derived values generally being smaller. Both systems give stable and consistent estimates of the areas of low temperature at 50 hPa. On the basis of data quality alone, it is not possible to recommend either analysis system in preference to the other for studies of the coldness of the polar stratosphere. ¿ 1999 American Geophysical Union