Abstract. Atmospheric profiles retrieved from GNSS (Global Navigation Satellite System) radio occultation (RO) measurements are increasingly used to validate other measurement data. For this purpose it is important to be aware of the characteristics of RO measurements. RO data are frequently compared with vertical reference profiles, but the RO method does not provide vertical scans through the atmosphere. The average elevation angle of the tangent point trajectory (which would be 90° for a vertical scan) is about 40° at altitudes above 70 km, decreasing to about 25° at 20 km and to less than 5° below 3 km. In an atmosphere with high horizontal variability we can thus expect noticeable representativeness errors if the retrieved profiles are compared with vertical reference profiles. We have performed an end-to-end simulation study using high-resolution analysis fields (T799L91) from the European Centre for Medium-Range Weather Forecasts (ECMWF) to simulate a representative ensemble of RO profiles via high-precision 3-D ray tracing. Thereby we focused on the dependence of systematic and random errors on the measurement geometry, specifically on the incidence angle of the RO measurement rays with respect to the orbit plane of the receiving satellite, also termed azimuth angle, which determines the obliquity of RO profiles. We analyzed by how much errors are reduced if the reference profile is not taken vertical at the mean tangent point but along the retrieved tangent point trajectory (TPT) of the RO profile. The exact TPT can only be determined by performing ray tracing, but our results confirm that the retrieved TPT – calculated from observed impact parameters – is a very good approximation to the "true" one. Systematic and random errors in RO data increase with increasing azimuth angle, less if the TPT is properly taken in to account, since the increasing obliquity of the RO profiles leads to an increasing sensitivity to departures from horizontal symmetry. Up to an azimuth angle of 30°, however, this effect is small, even if the RO profiles are assumed to be vertical. For applications requiring highest accuracy and precision it is advisable to exclude RO profiles with ray incidence angles beyond an azimuth of 50°. Errors in retrieved atmospheric profiles decrease significantly, by up to a factor of 2, if the RO data are exploited along the retrieved TPT. The tangent point trajectory of RO profiles should therefore be exploited whenever this is possible.