THE basic mechanical process responsible for earthquakes and faulting is not known. The intuitive notion of frictional slip on faults between elastic crustal blocks cannot be reconciled with laboratory measurements of the strength of rocks(1), field observations of heat flow(2,3) and stress orientation(4) around the San Andreas fault in California, and seismological estimates of the energy radiated by earthquakes(5). The weakening of large faults by elevated pore-fluid pressure(6) has been suggested as one solution to this paradox, but places severe constraints on the hydrological conditions of the faults concerned, Here I propose an alternative model for earthquake mechanics, in which the crust is treated as a system of many interlocking blocks divided by many faults(7), The model combines this random granular structure with simple, deterministic mechanical interactions, Numerical simulations of the deformation of an aggregate of rough grains under compressive stress show earthquake-like elastodynamic failures without frictional heat production, and substantial rotation of stresses across shear zones, which mimics field observations, There remain problems of scale in comparing these simulations with nature, but a seismological test of the model may be possible.