A method has been developed for the retrieval of mesospheric temperatures in the 65--90 km altitude range from satellite observations made by the Wind Imaging Interferometer (WINDII) aboard the Upper Atmosphere Research Satellite (UARS). Retrieved temperatures are derived from Rayleigh scattered sunlight observed in a wavelength band centered at 553 nm. Integrated line-of-sight radiance observations are inverted to tangent height volume scattering profiles, which are proportional to atmospheric density. From these, absolute temperature profiles are calculated using a technique derived from established Rayleigh lidar retrieval methods assuming that the atmosphere is in hydrostatic equilibrium and that it obeys the ideal gas law. Sources of error have been identified and typical temperature uncertainty values for individual profiles are determined to be <2.5%, 5.5%, and 13% for altitudes of 70 km, 80 km, and 90 km, respectively. A thorough comparison of the derived WINDII temperatures is performed against a number of ground-based and satellite measurements, including ground-based lidar, falling spheres, the High Resolution Doppler Imager observations aboard UARS, and against common atmospheric models. The data consist of spring equinox observations in March and April 1992/1993, summer solstice data in July/August 1992/1993, fall equinox data in September/October 1992, and winter solstice data in December 1992/1993 and January 1993/1994. The results of the comparisons show that WINDII temperatures are in reasonable to excellent agreement with a number of established temperature studies. In particular, July Northern Hemisphere monthly averaged temperatures show that characteristics of the mesopause obtained by WINDII are in very good agreement with other measurements. This good agreement with other established data sets and a determination of the error bounds of our recovered temperatures have shown that WINDII data can be used to confidently derive near-global temperatures of the upper mesosphere between 65 and 90 km. Above 90 km the errors increase, and systematic differences may arise with other measurements. ¿ 2001 American Geophysical Union