The molecular beam Fourier-transform microwave spectrum of pinacolone (methyl tert-butyl ketone) has been measured in several regions between 2 and 40 GHz. Fits of the assigned spectrum using several computer programs based on different models for treating torsion-rotation interaction lead to the conclusion that no existing program correctly captures the internal dynamics of this molecule. Quantum chemical calculations at the MP2/6-311++G(d,p) level of theory indicate that this molecule does not have a plane of symmetry at equilibrium, and that internal rotation of the light methyl group induces a large oscillatory motion of the heavy tert-butyl group from one side of the Cs configuration to the other. This effect has been modeled for J = 0 levels by a relatively simple two-top torsional Hamiltonian, where the magnitudes of the strong coupling terms between the tops are determined directly from the ab initio two-dimensional potential surface. A plot of the resultant 0A, 0E, 1E, 1A torsional levels on the same scale as a one-dimensional potential curve along the zigzag path connecting the six (unequally spaced) minima bears a striking resemblance to the 1:2:1 splitting pattern of the A, E, E, B levels of an internal rotation problem with a six-fold barrier. It seems likely that rotational transitions within the 1E and 1A torsional levels are the cause of the roughly 50% of the spectrum that remains unassigned after all predicted transitions within the 0A and 0E torsional levels are removed. However, a much more complete measurement campaign and some new torsion-rotation theory will be needed to verify this hypothesis.