A new wave retrieval method for the ERS synthetic aperture radar (SAR) wave mode is presented. The new algorithm, named semiparametric retrieval algorithm (SPRA), uses the full nonlinear mapping relations as proposed by Hasselmann and Hasselmann <1991>. It differs from previous retrieval algorithms in that it does not require a priori information on the sea state. Instead, it combines the observed SAR spectrum with the collocated wind vector from the ERS scatterometer to make an estimate of the wind sea spectrum. The residual signal in the SAR spectrum is interpreted as swell. The method has been validated by collocating over 5 years of SAR wave mode observations with spectral buoy measurements at 11 locations on the open ocean. For wave components longer than 225 m, the standard deviation between the retrieved spectra and buoy observations is 0.41 m, which corresponds to a relative RMS error of 29%. About 10% of the observed SAR spectra were rejected, in particular in light wind conditions when nonwave features such as those caused by slicks dominated the imagette. The bias and scatter in the results obtained under light wind conditions could be reduced by introducing a wind-dependent tilt modulation. This wind-dependent tilt formulation is derived from the empirical CMOD4 relation between the wind vector, the incidence angle, and the radar backscatter for the ERS scatterometer. ¿ 2000 American Geophysical Union