Seismic attenuation measurements provide a powerful tool for sampling mantle properties.Laboratory experiments provide calibrations at seismic frequencies and mantle temperatures for dry meltfree rocks, but require $10 2 210 3 extrapolations in grain size to mantle conditions; also, the effects of water and melt are not well understood. At the same time, body wave attenuation measured from dense broadband arrays provides reliable estimates of shear wave attenuation (Q 21 S ), affording an opportunity for calibration. We reanalyze seismic data sets that sample arc and back-arc mantle in Central America, the Marianas, and the Lau Basin, confirming very high attenuation (Q S $ 25-80) at 1 Hz and depths of 50-100 km. At each of these sites, independent petrological studies constrain the temperature and water content where basaltic magmas last equilibrated with the mantle, 1300-1450 C. The Q S measurements correlate inversely with the petrologically inferred temperatures, as expected. However, dry attenuation models predict Q S too high by a factor of 1.5-5. Modifying models to include effects of H 2 O and rheologydependent grain size shows that the effects of water-enhanced dissipation and water-enhanced grain growth nearly cancel, so H 2 O effects are modest. Therefore, high H 2 O in the arc source region cannot explain the low Q S , nor in the back arc where lavas show modest water content. Most likely, the high attenuation reflects the presence of melt, and some models of melt effects come close to reproducing observations. Overall, body wave Q S can be reconciled with petrologic and laboratory inferences of mantle conditions if melt has a strong influence beneath arcs and back arcs.