According to the classical theory, earthquakes would be caused by fracture, or the release of static friction, followed by sliding between the fault walls and a consequent drop of stress around the fault. At a depth of 600–700 km, where the pressure is about 200,000 bars, the lowest possible estimate of the coefficient of friction (about 1) would demand principal stress differences of the order of 400,000 bars for overcoming the friction of the fault wall and producing a drop in seismic stress. Such high stresses cannot be assumed to exist, if only because the yield stress of the asthenospheric rocks can hardly have an order of magnitude exceeding a few tens or at most hundreds of bars. This estimate is obtained by extrapolation of the observed creep behavior of hard materials with the assumption that the effect of a hydrostatic pressure is similar to that of a correspondingly increased molecular cohesive pressure; its order of magnitude agrees with that of the stress drop estimated from the energies released in earthquakes. A detailed discussion of this difficulty shows that earthquakes, except at focal depths less than about 5–10 km, cannot arise in the manner implied in the classical theory. The only plausible alternative available at present is that they are due to an instability of plastic deformation (creep) such as gives rise to slip bands, Lüders bands, the Hanson-Wheeler creep-deformation bands, and many other similar phenomena. If creep produces structural changes that accelerate further creep, the deformation concentrates gradually into thin layers in which high flow rates can develop, and finally even shear melting may occur by the heat development due to plastic deformation. Such a mechanism would explain the sequence structure of earthquakes (fore- and aftershocks). If fracture and frictional sliding are impossible and faulting can occur only by the gradual concentration of creep deformation into thin zones, the stress concentration around the edges of a fault cannot propagate this fault immediately over the entire stressed volume, as would be the case with faulting due to fracture. The fault can extend only after the occurrence of a certain amount of creep leading to progressive strain concentration in the region put under increased stress by the preceding faulting. The first faults in a sequence are likely to be small; after several preparatory faults have thrown a high stress upon a larger volume, however, extensive faulting with larger shocks may develop. Toward the end of the sequence the intensity of the shocks is likely to decrease again by depletion of stress. Since creep instability seems at present the only feasible mechanism of deep and intermediate-depth earthquakes, the fact that it is a characteristic of crystalline materials indicates that the seismogenic parts of the Earth’s mantle are substantially crystalline.