I describe an algorithm for retrieving geophysical parameters over the ocean from special sensor microwave/imager (SSM/I) observations. This algorithm is based on a model for the brightness temperature TB of the ocean and intervening atmosphere. The retrieved parameters are the near-surface wind speed W, the columnar water vapor V, the columnar cloud liquid water L, and the line-of-sight wind WLS. I restrict my analysis to ocean scenes free of rain, and when the algorithm detects rain, the retrievals are discarded. The model and algorithm are precisely calibrated using a very large in situ database containing 37,650 SSM/I overpasses of buoys and 35,108 overpasses of radiosonde sites. A detailed error analysis indicates that the TB model rms accuracy is between 0.5 and 1 K and that the rms retrieval accuracies for wind, vapor, and cloud are 0.9 ms-1, 1.2 mm, and 0.025 mm, respectively. The error in specifying the cloud temperature will introduce an additional 10% error in the cloud water retrieval. The spatial resolution for these accuracies is 50 km. The systematic errors in the retrievals are smaller than the rms errors, being about 0.3 ms-1, 0.6 mm, and 0.005 mm for W, V, and L, respectively. The one exception is the systematic error in wind speed of -1.0 ms-1 that occurs for observations within ¿20¿ of upwind. The inclusion of the line-of-sight wind WLS in the retrieval significantly reduces the error in wind speed due to wind direction variations. The wind error for upwind observations is reduced from -3.0 to -1.0 ms-1. Finally, I find a small signal in the 19-GHz, horizontal polarization (hpol)TB residual ΔTBH that is related to the effective air pressure of the water vapor profile. This information may be of some use in specifying the vertical distribution of water vapor. ¿ 1997 American Geophysical Union