The influence of intermittent convection on surface-layer stress estimates during the GARP Atlantic Tropical Experiment (GATE) is described. A negative correlation between the drag coefficient (C,) and the wind speed (0) is found when short averaging periods are used. Well-defined, discrete events produce this negative correlation, and these events are shown to correspond to the passage of convective plumes. Constraints on averaging times necessary to obtain reasonable stress estimates using the bulk method are discussed.Conditional sampling is used to produce average values of dissipation (E), wind speed (0). and virtual temperature (F,,) for each high turbulent intensity event, and for the quiescent periods in between. Such statistics indicate that the highly turbulent states coincide with the presence of plumes and account for the negative correlation between C, and I? Some of these statistics are also stability dependent.The probability distributions of the dissipation rate are bimodally log-normal which suggests that turbulence generated at two different heights is being sampled. This, along with other results of this paper, support a picture of a boundary layer which is dominated by vertical exchange.