In this study, we first estimate the representativeness errors for typical NWP models including the Weather Research and Forecast (WRF) model and the ECMWF model and investigate how to reduce the errors by a priori filtering of the raw RO bending angle observations. Then we explore the systematic errors of the analyses induced by the representativeness errors and the benefits of the a priori filtering by performing assimilation experiments of the raw and filtered bending angles with the WRF model using a sequential ensemble filtering technique. It is found that random representativeness errors can reach up to 8% in the moist lower troposphere, but a priori filtering of the raw bending angles can reduce the errors substantially to less than 2%. The large representativeness errors associated with raw bending angle data can introduce noticeable systematic errors in the analysis of water vapor in the lower troposphere and temperature in the upper troposphere. Assimilation of the a priori filtered bending angle observations substantially reduces such systematic errors and improves the analyses. We also find that the induced systematic errors are sensitive to the physical parameterizations of the model used in the assimilation. The results should help improve assimilation of current and, particularly the coming COSMIC-2 RO bending angle observations.