In July 1982 two detailed sections of density and absolute velocity were taken across the Gulf Stream northeast of Cape Hatteras to conduct an accurate test of geostrophy in a strong current. The sections, which were taken about 4 days apart, were each completed within 48 hours, using one ship to make closely spaced (12 km) conductivity-temperature-depth measurements to 2000 m, and a second ship to simultaneously take Pegasus absolute velocity profiles to the ocean bottom. The Gulf Stream path and curvature were also surveyed. The dynamically inferred velocity profiles were made absolute by matching their velocities to the Pegasus profiles at 2000 m. The geostrophic method (properly referenced) underestimated the observed velocities by 10--25 cm s-1 in the core of the current above 500 m where speeds exceed 150 cm s-1. The difference is a factor of 2 larger than the sampling and measurement errors in corresponding parts of the current, estimated to be 5--10 cm s-1 in the strong current and 2--5 cm s-1 elsewhere. We can account for the supergeostrophic currents quite effectively by including the centripetal acceleration (from the path curvature) in the momentum equation. In this case the differences in the current core decrease to less than 5--10 cm s-1, well within the uncertainties of the method. The transport above 2000 m is less sensitive to curvature effect; it agrees to within 3 Sv before and 1 Sv after correcting for curvature, or approximately 2% of the total 93-Sv transport. The deep velocity field below 1000 m had significant changes (10--20 cm s-1) in structure, and even reversals, in the 4 days between the sections, with cross-stream scales of 50--100 km. This deep variability, attributed to topographic Rossby waves, introduces more than 10 Sv uncertainty in defining the total volume transport of the Gulf Stream. ¿ American Geophysical Union 1989