The heavy jet mass distribution in e + e − collisions is computed to next-to-next-tonext-to leading logarithmic (N 3 LL) and next-to-next-to leading fixed order accuracy (NNLO). The singular terms predicted from the resummed distribution are confirmed by the fixed order distributions allowing a precise extraction of the unknown soft function coefficients. A number of quantitative and qualitative comparisons of heavy jet mass and the related thrust distribution are made. From fitting to ALEPH data, a value of α s is extracted, α s (m Z ) = 0.1220 ± 0.0031, which is larger than, but not in conflict with, the corresponding value for thrust. A weighted average of the two produces α s (m Z ) = 0.1193 ± 0.0027, consistent with the world average. A study of the non-perturbative corrections shows that the flat direction observed for thrust between α s and a simple non-perturbative shape parameter is not lifted in combining with heavy jet mass. The Monte Carlo treatment of hadronization gives qualitatively different results for thrust and heavy jet mass, and we conclude that it cannot be trusted to add power corrections to the event shape distributions at this accuracy. Whether a more sophisticated effective field theory approach to power corrections can reconcile the thrust and heavy jet mass distributions remains an open question.