Global tropospheric and stratospheric temperature trends and variations are studied in this paper using NCEP-NCAR reanalysis, ECMWF reanalysis ERA-40, and the Comprehensive Aerological Reference Data Set (CARDS). CARDS dataset is based on daily (up to four per day) radiosonde observations collected from over 20 data sources and range from 1940 to 2000. In this study 155 stations globally distributed are considered. A global comparison of temperature trends from NCEP-NCAR and ECMWF reanalyses in different layers (850-300mb, 300-100mb and 100-50mb) is presented, showing the different regional tendencies. CARDS stations are selected based on quality, consistency and sufficiently long-term of the temperature time series. These time series are compared, in every layer, to their equivalent in the closest NCEP-NCAR and ECMWF grid points by computing annual cycle, variability in different time scales, bias and trends. The low-frequency component of the time series is used to estimate the temperature trends. Statistical tests are used to evaluate the significance of the results. An assessment is made of the sampling issues associated with using the CARDS data set to infer global temperature trends, and also of the utility of the reanalyses for assessing temperature trends in data sparse regions.