The nucleation of earthquakes on weak faults is still poorly understood. Favored models for weakening have invoked the presence of high-pressure fluids contained within the fault. Here, from detailed field mapping, permeability measurements, and modeling, we assess how pressurized fluids may be impounded by layers of low-permeability phyllosilicate- rich fault gouge. Constraints are made on the permeability anisotropy, fluid-flow pathways, and fluid-loss rates from a large transcurrent fault zone. It is concluded that phyllosilicate-rich fault gouges having permeabilities ranging from 10−17 to 10−21 m2 and cumulative layer thicknesses of ∼50–200 m (from field observations) need fluid fluxes at the base of the brittle part of a large transcurrent fault (assumed to extend from 0 to 15 km depth) of ∼0.1 m3·m−2·yr−1. Furthermore, permeability anisotropy must exceed three orders of magnitude for overpressures to develop. This finding implies that vertical permeability must be enhanced by relatively permeable inclusions of fractured protolith within the fault zone, enclosed by walls of low-permeability gouge. For cross-zone flow, the lateral continuity of the phyllosilicate-rich fault-gouge layers provides effective barriers to fluid flow both into and out of the fault zone. This geometry restricts the origin of fluids in large fault zones as it favors fluids derived from deep sources. However, the fluid flux over time of mantle-derived CO2 and water produced from dehydration reactions has been estimated and appears inadequate. Given the frictional characteristics of phyllosilicate-rich fault gouge, small earthquakes and fault creep are the most likely modes of failure, facilitated by the buildup of high-pressure fluids in cells, producing spatial weakening of the fault.