Preliminary measurements of the stress orientation at a depth of 2 km in the Cajon Pass Scientific Drillhole, about 3.6 km from the San Andreas fault in southern California, indicate that the local direction of maximum compression is nearly normal to the fault zone. The measurements have been interpreted to indicate that the regional orientation of the maximum compression is normal to the fault, and taken as evidence for a very weak fault. The orientation of the regional maximum compression expected from plate tectonic arguments is about 66¿ NE from the strike of the fault. Geodetic data indicate that the orientation of maximum compressive strain rate is about 43¿ NE from the strike of the fault, and show nearly pure right-lateral shear acting parallel to the fault. These apparent conflicts in the inferred orientation of the axis of maximum compression may be explained in part by a model in which the fault zone is locked over a depth interval in the range of 2--5 to 15 km, but is very weak above and below that interval. An analytic solution for this model predicts orientations of the principal stresses that are nearly parallel or normal to the fault, for observations made close to the fault and above and below the locked zone, but an orientation consistent with plate tectonic estimates at depths corresponding to that of the locked interval. This solution does require, however, a few mm/yr of creep at the surface on the San Andreas or nearby sub-parallel faults (such as the San Jacinto), which has not yet been observed, or a shallow zone near the faults of distributed deformation.