We predict plate motions from a comprehensive inversion of theoretical estimates of tectonic forces in order to evaluate the relative importance of these and the uncertainties of such models. Plate-driving forces from the mantle are calculated using global flow models that are driven by tomography and subduction-derived density fields. Observed and predicted plate velocities agree well for a variety of models, leading to varied conclusions about the relative importance of forces. The dominance of the subduction related density pattern in the mantle is confirmed; it appears that P wave models do not satisfactorily image all of the slab-associated anomalies in the upper mantle. Furthermore, lower mantle structure always improves the plate motion fit with respect to models that are based on upper mantle anomalies and lithospheric thickening only. We show that the average torques from the lower mantle scale with the radial flow through the 660-km phase transition; the amplitude of the lower mantle torques will be significant for a range of models if there is mass flux through 660 km. We also evaluate parameterized edge forces and find that the additional inclusion of such torques does not significantly improve the model fit. The main reason for the nonuniqueness of the inversions is plate boundary geometry since all plate motions are dominated by the trench-ridge system, and plates move from ridges to trenches.