Wave spectra that are retrieved from ERS-1/2 synthetic aperture radar (SAR) wave mode observations with two different algorithms are validated against 6 years of buoy observations. The Max-Planck Institut f¿r Meteorologie (MPIM) algorithm, which runs operationally at the European Center for Medium-Range Weather Forecasts (ECMWF), is found to deteriorate the quality of the WAM spectrum which is used as a first guess. The Semi-Parametric Retrieval Algorithm (SPRA) does not use a first-guess spectrum. For wavelengths which are observed by the SAR, it has a skill comparable to WAM. Several causes for the poor performance of the MPIM scheme are suggested. First, despite the fact that the SAR generally does not resolve the wind sea peak, the MPIM scheme allows for independent adjustment of its energy and peak frequency. Second, by using the quasi-linear approximation in the inversion, the scheme is inclined to interpret the SAR signal at low wave numbers as swell, whereas often it is generated by waves at higher wave numbers via nonlinearities in the SAR mapping. Third, the MPIM scheme is not able to adjust the spectral width of wave systems. The SPRA scheme retrieves swell information only up to a 180¿ directional ambiguity, and the SPRA retrievals often contain a spectral gap between the shortest waves observed by the SAR and the parameterized wind sea. In conclusion, the retrieval scheme performing best is the SPRA scheme, which has an accuracy comparable to WAM model output for the longer-swell waves. ¿ 2001 American Geophysical Union