Over a 2-month period the seabed drag coefficient C100 at a site in the eastern Irish Sea (26-m water depth) had a mean value of 0.0025, which is the value typically assumed, but was highly variable. The objective of the analysis was to unravel the causes of the variability in C100. An error analysis revealed that a considerable part of the variability could be due simply to random errors in the estimates of friction velocity, from which C100 was deduced, but the error bounds did not enclose all of the data. Wave-current interaction was ruled out as the principal cause of the variability because waves were small and there was no noticeable correlation between C100 and ratio of wave-orbital speed to tidal-current speed. Classifying the boundary-layer flow into smooth-turbulent, transitional, and rough-turbulent regimes implied that flow-regime transitions could explain much of the variability, however, inspection of the data revealed a contradiction that made that explanation unlikely: the classification implied that the boundary-layer flow was rough turbulent during neap tides but was smooth turbulent under peak spring-tide currents. The classification scheme was based on the assumption of neutral stability, which according to the stability parameter z/L, was violated for a considerable portion of the time. Boundary-layer stratification by suspended sediment was found to provide a coherent explanation for drag coefficient variability; when z/L0.03, the buoyancy flux due to the suspended sediment caused C100 to vary widely. The analysis of the stratified-flow dynamics includes the development of an equation (correct to second order in z/L, where z/L is small) for estimating bed shear stress from the velocity spectrum, which extends the applicability of the well-known ''inertial-dissipation'' method into the region z/L>0.03. ¿ American Geophysical Union 1995