Smoothed records of ice drift, surface wind and upper ocean currents at four manned stations of the 1975–76 AIDJEX experiment in the central Arctic have been analyzed to provide a statistical relationship between stress at the ice-ocean interface and ice-drift velocity during a 60-day period when the ice, was too weak to support internal forces. Using interfacial stress calculated from a balance with air stress and Coriolis force on the ice column for times longer than the inertial period, logarithmic linear regression of the stress-velocity samples provided the relation τ = 0.010V1.78, where τ is the magnitude of interfacial stress and V the ice speed relative to the geostrophic current in the ocean. This result is statistically indistinguishable from predictions of a numerical model adapted from Businger and Arya (1974) with surface roughness Z0 = 10 cm. Essential features of the model are dynamic scaling by u*, u*2 and u*/f for velocity, kinematic stress and length, with exponential attenuation of a linear dimensionless eddy viscosity, viz., K* = −kξeϵ1ξ, where ξ = fz/u* and k is von Kaa's constant. Currents measured 2 m below the ice confirmed the shape of the τ vs V curve and provided an estimate of the angle between surface stress and velocity. The model was used to qualitatively estimate the effect of a pycnocline at 25 m on surface characteristics. The observed behavior when stratification at that level was most pronounced tended toward slightly higher drag at higher speeds, which is qualitatively consistent with the model results.