We use spin-transfer-driven ferromagnetic resonance (ST-FMR) to measure the spin-transfer torque vector τ in MgO-based magnetic tunnel junctions as a function of the offset angle between the magnetic moments of the electrodes and as a function of bias, . We explain the conflicting conclusions of two previous experiments by accounting for additional terms that contribute to the ST-FMR signal at large . Including the additional terms gives us improved precision in the determination of τ(V), allowing us to distinguish among competing predictions. We determine that the in-plane component of dτ/dV has a weak but nonzero dependence on bias, varying by 30%–35% over the bias range where the measurements are accurate, and that the perpendicular component can be large enough to be technologically significant. We also make comparisons to other experimental techniques that have been used to try to measure τ(V).