Within the ERA-40 reanalysis project, an analysis feedback (AF) dataset has been generated. The AF contains differences between ERA-40 6h-background forecasts and observations as well as differences between the final analyses and observations. By investigating several radiosondes over Central Europe it is demonstrated that this dataset is appropriate for checking the temporal homogeneity of individual radiosonde records. The study corroborates earlier results by others that radiosonde records need to be corrected before they can be used as input for a temporally homogeneous reanalysis dataset. Intercomparisons with many radiosonde stations indicate that not only the radiosonde series but also the time series of the background forecasts and of the final analyses contain inhomogeneities. The warming trend from 6h-background forecasts in 700 hPa over Central Europe is found to be much weaker than trends from radiosondes and independent homogenized Alpine summit station observations. A relatively large break in the 6h-background forecast series in January 1973, when the first satellite data have become available, has been identified as the main reason for the underestimated trend. In this preliminary study only some breaks in the observation data as well as in the ERA-40 data have been detected as examples for demonstrating the potential of the AF dataset. No attempt has yet been made to correct the breaks using the AF data.