Earthquake mechanisms based on frictional instabilities are widely accepted for relatively shallow earthquakes. Such mechanisms, in unmodified form, are highly unlikely for deep earthquakes because of the expected high frictional shear stresses at depths of 300--700 km. Several alternative mechanisms have been proposed. Here, I discuss a new type of shear instability observed in experiments on ordinary polycrystalline ice and polycrystalline tremolite by my colleagues and me. These shear instabilities lead to rapid stress drops and sound emissions and have the following characteristics; (1) A single sharp shear zone forms in the maximum shear stress orientation. (2) Confining pressure does not suppress the instability and, under some conditions, rasing confining pressure Pc lowers the differential stress &sgr;f at which it occurs. (3) Stress drops can occur repeatedly after reloading at about the same differential stress. (4) The mean stress vs. temperature conditions at which the instabilities occur in ice and tremolite plate them in the fields of thermodynamic stability of ice II and diopside+talc, respectively. Although I have not as yet confirmed the existence of these high pressure phases in the shear zones, transformation in the shear zone is implicated because such shear instabilities do not occur in similar materials in which polymorphic phase changes do not occur. Reconstructive polymorphic phase changes localized along shear zones may be a general phenomenon in crystalline materials stressed non-hydrostatically near transition boundaries. Moreover, by analogy, the reconstructive phase transformations that are thought to occur in deeply subducted lithosphere may lead to localized shear instabilities manifested by deep earthquakes and produce the observed double-couple first-motion distribution. Finally, if this is the earthquake source mechanism in deeply subducted lithosphere, then earthquakes should shut off when all of the polymorphic phase changes have occurred. This may explain the maximum depth of earthquakes at about 680 km. ¿ American Geophysical Union 1987