[1] New geophysical data collected at the Aden-Owen-Carlsberg (AOC) triple junction between the Arabia, India, and Somalia plates are combined with all available magnetic data across the Gulf of Aden to determine the detailed Arabia-Somalia plate kinematics over the past 20 Myr. We reconstruct the history of opening of the Gulf of Aden, including the penetration of the Sheba Ridge into the African continent and the evolution of the triple junction since its formation. Magnetic data evidence three stages of ridge propagation from east to west. Seafloor spreading initiated ∼20 Myr ago along a 200 kmlong ridge portion located immediately west of the Owen fracture zone. A second 500 kmlong ridge portion developed westward up to the Alula-Fartak transform fault before Chron 5D (17.5 Ma). Before Chron 5C (16.0 Ma), a third 700 km-long ridge portion was emplaced between the Alula-Fartak transform fault and the western end of the Gulf of Aden (45°E). Between 20 and 16 Ma, the Sheba Ridge propagated over a distance of 1400 km at an extremely fast average rate of 35 cm yr −1. The ridge propagation resulted from the Arabia-Somalia rigid plate rotation about a stationary pole. Since Chron 5C (16.0 Ma), the spreading rate of the Sheba Ridge decreased first rapidly until 10 Ma and then more slowly. The evolution of the AOC triple junction is marked by a change of configuration around 10 Ma, with the formation of a new Arabia-India plate boundary. Part of the Arabian plate was then transferred to the Indian plate.

S U M M A R YWe derive the first chronologically detailed model of Eurasia-North America plate motion since 20 Ma from ship and airplane surveys of the well-expressed magnetic lineations along this slowly spreading plate boundary, including previously unavailable dense Russian magnetic data from the southern Reykjanes Ridge and northern Mid-Atlantic ridge near the Charlie Gibbs fracture zone. From more than 7000 crossings of 21 magnetic anomalies from Anomaly 1n (0.78 Ma) to Anomaly 6n (19.7 Ma), we estimate best-fitting finite rotations and realistic uncertainties. Linear regressions of total opening distances versus their reversal ages at different locations along the plate boundary show that reversal boundaries are shifted systematically outwards from the spreading axis with respect to their idealized locations, with the outward shift ranging from more than 5 km between Iceland and the Charlie Gibbs fracture zone to ∼2 km elsewhere. This outward displacement, which is a consequence of the finite zone of seafloor accretion, degrades estimates of the underlying plate motion and is thus removed for the ensuing kinematic analysis. The corrected plate motion rotations reveal surprising, previously unrecognized features in the relative motions of these two plates. Within the uncertainties, motion was steady from 20 to 8 Ma around a pole that was located ∼600 km north of the present pole, with seafloor spreading rates that changed by no more than 5 per cent (1 mm yr −1 ) along the Reykjanes Ridge during this period. Seafloor spreading rates decreased abruptly by 20 ± 2 per cent at 7.5-6.5 Ma, coinciding with rapid southward migration of the pole of rotation and a 5 • -10 • counter-clockwise change in the plate slip direction. Eurasia-North America plate motion since 6.7 Ma has remained remarkably steady, with an apparently stationary axis of rotation and upper limit of ±2 per cent on any variations in the rate of angular rotation during this period. Based on the good agreement between seismotectonic constraints on present deformation in northeast Asia and directions of motion that are predicted by our 6.7 Ma to present pole, we hypothesize that motion has remained steady to the present and attempt to test this hypothesis with published GPS estimates for Eurasia-North America motion. We find, however, that GPS estimates that are tied to recent versions of the international geodetic reference frame and rely principally on station velocities from Europe give implausible estimates of recent motion, with the most recently published GPS model predicting convergence along the southern Gakkel Ridge and in the Laptev Sea, where seafloor spreading occurs. An alternative GPS estimate that is not tied to the international terrestrial reference frame and employs GPS station velocities from northeastern Asia is marginally consistent with our 6.7-0 Ma motion estimate.