Error propagation from winds to ocean models was numerically investigated using the Princeton Ocean Model (POM) for the South China Sea with 20-km horizontal resolution and 23 &sgr; levels conforming to a realistic bottom topography during the lifetime of tropical cyclone Ernie (November 4--18, 1996). Numerical integration was divided into preexperimental and experimental stages. The preexperiment phase generates the initial conditions on November 1 for the sensitivity experiment. During the experimental stage the POM was integrated from November 1 to 30, 1996 under National Centers for Environmental Prediction (NCEP) reanalyzed surface fluxes along with two surface wind data sets, namely, the daily averaged interpolated NASA scatterometer winds and the NCEP winds. The relative root-mean-square differences fluctuate from 0.5 to 1.0 for winds, 0.25 to 0.7 for surface elevations, 0.47 to 1.02 for surface currents, and 0 to 0.23 for surface temperatures. This indicates that the model has less uncertainty overall than the wind fields used to drive it, which in turn suggests that the ocean modeling community may progress without waiting for the atmospheric modelers to build the perfect forecast model. ¿ 1999 American Geophysical Union