The Michelson Interferometer for Passive Atmospheric Sounding (MIPAS) onboard the Envisat satellite provides temperature and various gas profiles from limb-viewing midinfrared emission measurements. The stratospheric temperatures retrieved at the Institut f¿r Meteorologie und Klimaforschung (IMK) for September/October 2002 and October/November 2003 are compared with a number of reference data sets, including global radiosonde (RS) observations, radio occultation (RO) measurements of Global Positioning System (GPS) on German Challenging Minisatellite Payload (CHAMP) and Argentinean Satelite de Aplicaciones Cientificas-C (SAC-C) satellite, Halogen Occultation Experiment (HALOE) on the Upper Atmosphere Research Satellite (UARS), and the analyses of European Centre for Medium-Range Weather Forecasts (ECMWF) and Met Office (METO), United Kingdom. The data sets show a good general agreement. Between 10 and 30 km altitude the mean differences are within ¿0.5 K for the averages over the height interval and within ¿(1--1.5) K at individual levels for comparisons with RS, GPS-RO/CHAMP, and SAC-C, ECMWF, and METO. Between 30 and 45 km the MIPAS mean temperatures, averaged over the height region, are higher than ECMWF but lower than METO by ~1.5 K, while they differ by ¿0.5 K with respect to HALOE, with maximum discrepancies of ~2.5 K peaking around 35 km. Between 45 and 50 km, MIPAS temperatures show a low bias compared to HALOE, ECMWF, and METO with mean differences of -1 to -3 K and with a better agreement with HALOE. The large discrepancies between MIPAS and the analyses above 30 km likely suggest deficiency in the underlying general circulation models. The standard deviations vary between 2.5 and 3.5 K for individual data sets, with more than 70% being contributed from the expected variability of the atmosphere. Retrieved temperatures with accuracy of ~0.5--1 K after removing the atmospheric variability provide highly accurate knowledge to characterize our environment.