Brittle fracture and frictional sliding are impossible below a few tens of km yet earthquakes occur in subducting lithosphere to -700 km. Experimental work shows that at high pressure a small amount of low-viscosity "fluid" must be generated to enable shear failure and comparison with the earthquake distribution in the upper 300 km of subducting slabs strongly indicates that the method of earthquake initiation is dehydration of hydrous phases. In contrast, the earthquake distribution below 400 km shows no correlation with potential H <sub>2</sub> O-liberating reactions; indeed assuming that hydrous phases are present predicts earthquakes in places where they are not observed. Thus, the earthquake distribution suggests that H <sub>2</sub> O-releasing reactions do not take place in the mantle transition zone. In addition, metastable olivine has now been detected by seismic means in 2 subduction zones. Such metastable olivine can explain the earthquake distribution and also is incompatible with the presence of H <sub>2</sub> O, even in very small amounts. I conclude that subduction zones are essentially dry below 400 km.