Global Positioning System (GPS) measurements along the subduction zone of northern Sumatra (2 ¿S to 3 ¿N) reveal that the strain associated with the oblique convergence of the Australian plate with Eurasia is almost fully partitioned between trench-normal contraction within the forearc and trench-parallel shear strain within a few tens of kilometers of the Sumatran fault. Kinematic analyses of interplate earthquake slip vectors provide slip rates on the Sumatran fault within a few millimeters per year of GPS and geologic rates, giving us more confidence in the use of slip vectors for inferring slip partitioning elsewhere. The inferred slip rate on the Sumatran fault is ~1/3 less than the full margin parallel component of plate motion. An across-forearc rotation in the slip vectors suggests that the missing arc-parallel shear occurs seaward of the geodetic network, between the forearc islands and the trench. Simple finite element models are used to explore the conditions under which the change in the principal strain rate directions between the forearc and the arc region can occur. Modeling suggests that neither a preexisting strike-slip fault nor a zone of thermally induced lithospheric weakness in the overriding plate is needed for strain partitioning to occur. In general, forearc slivers form over the region of interplate coupling and are driven along strike by the basal shear. A volcanic arc can help the partitioning process by localizing the margin-parallel shear strain in the upper plate if its crust and mantle are weaker than its surroundings. Interplate slip vectors and geodetic results from Sumatra together suggest that the highest coupling on the plate boundary occurs beneath and seaward of the forearc islands, consistent with inferences about the rupture zones of great nineteenth century earthquakes there. The Sumatra example suggests that geodetic measurements of interseismic, margin-parallel shear strain at oblique convergent margins can be used to map the landward extent of the relatively high basal stress beneath the overriding plate if one can correct for strain localization caused by weak upper plate strike-slip faults. ¿ 2000 American Geophysical Union