Global Navigation Satellite System--based radio occultation (RO) measurements promise to become a valuable observational basis for climate research, numerical weather prediction, and atmospheric process studies, thanks to their unique combination of global coverage, high accuracy and vertical resolution, and long-term stability. Concerning the quality of the RO-derived atmospheric profiles, performance simulation studies and first data analysis results from the current CHAMP satellite mission are encouraging but also reveal weaknesses of present RO retrieval chains at high altitudes (above 30 km). This study aims at providing, first, understanding of these weaknesses and, second, mitigation by an advanced retrieval scheme. We first evaluated present state-of-the-art high-altitude RO retrieval algorithms. We found clear superiority for using statistical optimization involving background information over using exponential extrapolation and a sensitivity of RO retrieval products to biases in background information calling for sensible bias mitigation. Exploiting these findings, an advanced high-altitude retrieval scheme is presented, which focuses on minimizing residual retrieval biases in the upper stratosphere and thereby on optimizing the climate monitoring utility of retrieved profiles. Applied to a large ensemble of simulated occultation events, the advanced scheme proved to be effective. The scheme is currently under evaluation with real data from the CHAMP mission. It will then serve as part of a processing chain generating RO-based global climatologies of refractivity, geopotential height, temperature, and humidity based on RO data from CHAMP, SAC-C, GRACE and other future satellites carrying RO instruments.