Abstract. Radio occultation (RO) is a promising source of observation for weather and climate applications. However, the uncertainties arising from imperfect retrieval algorithms may weaken the overall confidence in the data and discourage their use. As an alternative approach of assessing the quality of RO data while avoiding the nuisance of retrieval errors, this study proposes to use minimally processed data (measurement) instead of derived RO data. This study compares measured phase paths with their model counterparts, simulated with an effective ray tracer for which the refractive indices along the complete ray path linking the transmitter and the receiver are realistically specified. The comparison of phase measurements with the European Centre for Medium-Range Weather Forecasts (ECMWF) data made in the observation space shows that the RO measurements are of sufficient accuracy to uncover regional-scale systematic errors in ECMWF's operational analysis and the 45-year reanalysis (ERA40), and to clearly depict the error growth of short-term ERA40 forecasts. In the southern hemispheric stratosphere, in particular, the RO measurements served as a robust reference against which both of the two analyses were significantly biased in opposite directions even though they were produced by the same center using virtually the same set of data. The measurement and ECMWF analyses showed a close agreement in the standard deviation except for the regions and heights that the quality of the ECMWF data is controversial. This confirms the high precision of RO measurements and also indicates that the main problem of the ECMWF analyses lies in their systematic error.