Theoretical calculations of global positioning system (GPS) signals reflected from rough sea surfaces are discussed and compared with the aircraft measurements. The theoretical model is based on the assumption that rough sea surfaces are composed of facets, and the reflection of GPS signals from the facets can be calculated on the basis of geometric optics. In order to determine the slopes of the sea surface facets, the statistical model of Cox and Munk <1954> for ocean surface slopes is used. Since much of the sea surface roughness observed is at scales much less than the GPS wavelength, the dependence of the mean square slopes on frequency is taken into consideration. Model results agree well with aircraft measurements: the differences between model results and observations are within the level of experimental errors. For calm sea surfaces (near-sea-surface wind speeds 7 m s-1), the correlation powers decrease their peak values to ~-4---8 dB, significantly widen time delay responses, and shift toward the delays corresponding to path lengths longer than those of specular points. Sensitivity tests show that there is considerable potential for current or advanced GPS receiving systems to estimate near-sea-surface wind speeds: the errors in the wind speed estimates could be smaller than 2 m s-1. For near-sea-surface wind directions, the maximum changes in the correlation powers of the GPS signals as a function of azimuth appear to be ~0.5 dB for current GPS receiving systems, which may be too small for wind direction estimation. New designs of the GPS receivers or a higher signal-to-noise ratio than that of current aircraft receivers may be needed. ¿ 1999 American Geophysical Union