Large gaps and inconsistencies remain in published estimates of Nubia–Somalia plate motion based on reconstructions of seafloor spreading data around Africa. Herein, we use newly available reconstructions of the Southwest Indian Ridge at ∼1-Myr intervals since 20 Ma to estimate Nubia–Somalia plate motion farther back in time than previously achieved and with an unprecedented degree of temporal resolution. At the northern end of the East African rift, our new estimates of Nubia–Somalia motion for six times from 0.78 Ma to 5.2 Ma differ by only 2 per cent from the rift-normal component of motion that is extrapolated from a recently estimated GPS angular velocity. The rate of rift-normal extension thus appears to have remained steady since at least 5.2 Ma. Our new rotations indicate that the two plates have moved relative to each other since at least 16 Ma and possibly longer. Motion has either been steady since at least 16 Ma or accelerated modestly between 6 and 5.2 Ma. Our Nubia–Somalia rotations predict 42.5 ± 3.8 km of rift-normal extension since 10.6 Ma across the well-studied, northern segment of the Main Ethiopian Rift, consistent with 40–50 km estimates for extension since 10.6 Myr based on seismological surveys of this narrow part of the plate boundary. Nubia–Somalia rotations are also derived by combining newly estimated Somalia–Arabia rotations that reconstruct the post-20-Ma opening of the Gulf of Aden with Nubia–Arabia rotations estimated via a probabilistic analysis of plausible opening scenarios for the Red Sea. These rotations predict Nubia–Somalia motion since 5.2 Myr that is consistent with that determined from Southwest Indian Ridge data and also predict 40 ± 3 km of rift-normal extension since 10.6 Ma across the Main Ethiopian Rift, consistent with our 42.5 ± 3.8 km Southwest Indian Ridge estimate. Our new rotations exclude at high confidence level previous estimates of 12 ± 13 and 123 ± 14 km for rift-normal extensions across the Main Ethiopian Rift since 10.6 Ma based on reconstructions of Chron 5n.2 along the Southwest Indian Ridge. Sparse coverage of magnetic reversals older than 16 Ma along the western third of the Southwest Indian Ridge precludes reliable determinations of Nubia–Somalia plate motion before 16 Ma, leaving unanswered the key question of when the motion between the two plates began.