Earth's deepest earthquakes occur as a population in subducting or previously subducted lithosphere at depths ranging from about 325 to 690 km. This depth interval closely brackets the mantle transition zone, characterized by rapid seismic velocity increases resulting from the transformation of upper mantle minerals to higher-pressure phases. Deep earthquakes thus provide the primary direct evidence for subduction of the lithosphere to these depths and allow us to investigate the deep thermal, thermodynamic, and mechanical ferment inside slabs. Numerical simulations of reaction rates show that the olivine ??? spinel transformation should be kinetically hindered in old, cold slabs descending into the transition zone. Thus wedge-shaped zones of metastable peridotite probably persist to depths of more than 600 km. Laboratory deformation experiments on some metastable minerals display a shear instability called transformational faulting. This instability involves sudden failure by localized superplasticity in thin shear zones where the metastable host mineral transforms to a denser, finer-grained phase. Hence in cold slabs, such faulting is expected for the polymorphic reactions in which olivine transforms to the spinel structure and clinoenstatite transforms to ilmenite. It is thus natural to hypothesize that deep earthquakes result from transformational faulting in metastable peridotite wedges within cold slabs. This consideration of the mineralogical states of slabs augments the traditional largely thermal view of slab processes and explains some previously enigmatic slab features. It explains why deep seismicity occurs only in the approximate depth range of the mantle transition zone, where minerals in downgoing slabs should transform to spinel and ilmenite structures. The onset of deep shocks at about 325 km is consistent with the onset of metastability near the equilibrium phase boundary in the slab. Even if a slab penetrates into the lower mantle, earthquakes should cease at depths near 700 km, because the seismogenic phase transformations in the slab are completed or can no longer occur. Substantial metastability is expected only in old, cold slabs, consistent with the observed restriction of deep earthquakes to those settings. Earthquakes should be restricted to the cold cores of slabs, as in any model in which the seismicity is temperature controlled, via the distribution of metastability. However, the geometries of recent large deep earthquakes pose a challenge for any such models. Transformational faulting may give insight into why deep shocks lack appreciable aftershocks and why their source characteristics, including focal mechanisms indicating localized shear failure rather than implosive deformation, are so similar to those of shallow earthquakes. Finally, metastable phase changes in slabs would produce an internal source of stress in addition to those due to the weight of the sinking slab. Such internal stresses may explain the occurrence of earthquakes in portions of lithosphere which have foundered to the bottom of the transition zone and/or are detached from subducting slabs. Metastability in downgoing slabs could have considerable geodynamic significance. Metastable wedges would reduce the negative buoyancy of slabs, decrease the driving force for subduction, and influence the state of stress in slabs. Heat released by metastable phase changes would raise temperatures within slabs and facilitate the transformation of spinel to the lower mantle mineral assemblage, causing slabs to equilibrate more rapidly with the ambient mantle and thus contribute to the cessation of deep seismicity. Because wedge formation should occur only for fast subducting slabs, it may act as a "parachute" and contribute to regulating plate speeds. Wedge formation would also have consequences for mantle evolution because the density of a slab stagnated near the bottom of the transition zone would increase as it heats up and the wedge tra