The presence of ocean surface waves has been observed to affect both the magnitude and direction of the wind stress. Here concurrent wind and wave data are employed to study their relationship. To help isolate the influence of the waves, the wind stress is broken into three frequency bands: low (frequencies below 0.06 Hz), corresponding to large-scale motions in the boundary layer at frequencies below any significant wave energy; middle (frequencies between 0.06 and 0.16 Hz), corresponding to the frequencies of the dominant swell and wind waves; and high (frequencies above 0.16 Hz), corresponding to waves too short to influence coherently the wind fluctuations at the anemometer site 8 m above the surface. Most often, the low band holds the most stress. The magnitude of the wind stress within the low band increases roughly with the square of the mean wind speed, the high band appears to increase with the wind speed to the fourth power, and the middle band exhibits varied dependence. The direction of the wind stress in the low band is closely tied to the mean wind direction. In contrast, the directions in the middle and high bands are influenced by the waves and can be significantly off the mean wind direction. The middle band is biased toward the direction of long-period swell, while the high band is biased toward the direction of short-period seas, which is closer to the wind direction. Thus it is mainly within the middle band that large deviations in stress versus wind magnitude and direction are found. To further isolate the influence of waves, a wave-correlated fraction of the wind stress is estimated using direct correlations between the surface elevation and wind fluctuations. Removing this wave-correlated stress from the total results in a residual stress that is better behaved: the magnitude of the residual stress in the middle band is modeled by a simple wind speed dependent drag coefficient, and the direction is very nearly aligned with the wind in both the middle and high bands. These results indicate that waves are indeed closely associated with the observed deviations from bulk formula stress estimates. They also suggest a new method by which to estimate the wind stress; namely, partitioning the stress into three separately modeled parts: a low-frequency stress, a high-frequency wave-correlated stress, and a high-frequency residual stress. ¿ 1998 American Geophysical Union