The CERES experiment studies the production of low-mass electron pairs in proton-proton, proton-nucleus and nucleus-nucleus collisions at the CERN-SPS. Particle identification and momentum measurement are solely based on two azimuthally symmetric RICH detectors. The off-line reconstruction of electron rings has to be preformed without prior knowledge of the cherenkov ring centers, a demand which is unique as compared to other RICH applications using the rings for particle identification or rejection only. During the last years the CERES collaboration has developed several algorithms to cope with the problems of ring recognition in a background-contaminated environment. Here we report on encouraging experiences with the robust ring fitting technique based on a reweighted least-squares method and compare its performance with previously used algorithms based on the minimization of summed Gaussian weights.