The rigid-plate hypothesis implies that oceanic lithosphere does not contract horizontally as it cools (hereinafter “rigid plate”). An alternative hypothesis, that vertically averaged tensional thermal stress in the competent lithosphere is fully relieved by horizontal thermal contraction (hereinafter “shrinking plate”), predicts subtly different azimuths for transform faults. The size of the predicted difference is as large as 2.44° with a mean and median of 0.46° and 0.31°, respectively, and changes sign between right-lateral (RL)-slipping and left-lateral (LL)-slipping faults. For the MORVEL transform-fault data set, all six plate pairs with both RL- and LL-slipping faults differ in the predicted sense, with the observed difference averaging 1.4° ± 0.9° (95% confidence limits), which is consistent with the predicted difference of 0.9°. The sum-squared normalized misfit, r, to global transform-fault azimuths is minimized for γ = 0.8 ± 0.4 (95% confidence limits), where γ is the fractional multiple of the predicted difference in azimuth between the shrinking-plate (γ = 1) and rigid-plate (γ = 0) hypotheses. Thus, observed transform azimuths differ significantly between RL-slipping and LL-slipping faults, which is inconsistent with the rigid-plate hypothesis but consistent with the shrinking-plate hypothesis, which indicates horizontal shrinking rates of 2% Ma−1 for newly created lithosphere, 1% Ma−1 for 0.1 Ma old lithosphere, 0.2% Ma−1 for 1 Ma old lithosphere, and 0.02% Ma−1 for 10 Ma old lithosphere, which are orders of magnitude higher than the mean intraplate seismic strain rate of ~10−6 Ma−1 (5 × 10−19 s−1).