Wind stress fields during summer 2000 and 2001 from the operational Eta mesoscale atmospheric forecast model and the QuikSCAT/SeaWinds scatterometer are compared for the coastal region west of Oregon and California, extending offshore to 130 W. A simple correction is proposed for systematic error in the model fields. The scatterometer measurements show regions of intensified stress near orographic features, consistent with previous measurements and results from high-resolution mesoscale model simulations. In these regions of orographic intensification, the moderate-resolution (32-km) Eta model systematically underpredicts the observed wind stress by a factor reaching 1/2 or less, and places the axis of the maximum wind farther offshore than do the scatterometer measurements. However, the temporal correlations between Eta-model and QuikSCAT wind stresses remain high (0.85-0.9) even in these regions of intensified stress, except in a narrow region adjacent to the coast, where the correlations decrease abruptly. These high correlations motivate the use of a complex linear regression model to reduce the systematic error, relative to QuikSCAT measurements, in the forecast-model coastal-zone wind stress. The regression model is calibrated using a split-sample validation method, applied for summer 2000 and summer 2001 data. The regression model produces a wind stress field on the nominal 25-km QuikSCAT grid from 40-km Eta output, and substantially reduces the Root Mean Square Error (RMSE) relative to QuikSCAT measurements. The RMSE of the regression-corrected Eta model is smaller than that of two high-resolution, nested, mesoscale dynamical models, implemented using standard procedures.