We use precise hypocenter patterns and focal mechanisms to investigate the presence or absence of a continuous strike-slip fault at depth connecting the San Bernardino strand of the San Andreas fault with the Coachella Valley segment of the Banning fault. We inverted 560,000 arrival times from 23,000 earthquakes (1981--1993) for high-quality hypocenters and three-dimensional P wave velocity structure in a 1¿ by 2¿ area centered on the San Gorgonio Pass. Cross-sectional plots of relocated earthquakes reveal an abrupt 5 to 7 km high step in the maximum depth of hypocenters. The step riser defines a near-vertical, locally curved surface that extends westerly more than 60 km from the Coachella Valley segment of the San Andreas fault to the San Jacinto fault. A hypothetical continuous vertical San Andreas fault through San Gorgonio Pass would cross the step at an oblique angle. We suggest that the step is the expression of the contact between different basement rock types juxtaposed by large-scale right-slip motion on the ancestral San Andreas fault. South of the step in Peninsular Ranges type basement (intrusives), brittle failure occurs down to about 20-km depth, while north of the step in San Bernardino type basement (Pelona schist), brittle failure occurs to only about 13-km depth. The step provides a piercing plane that should be offset about 3 km right laterally by an active, continuous, vertical San Andreas fault. Within the resolution of our mapping the step is not offset in this manner, implying either that there has not been a throughgoing vertical fault at depth, that a throughgoing fault has not experienced enough slip to offset the step, or that a throughgoing fault is not vertical and dips north over the top of the step. Hypocentral patterns and focal mechanisms indicate distributed deformation (thrust, normal, and strike-slip faulting) over a large volume in the San Gorgonio Pass region; there is no evidence of hypocenter or slip vector alignments that would indicate a throughgoing, continuous, near-vertical San Andreas fault. In summary, we find no evidence indicating a continuous fault at seismogenic depth connecting the San Bernardino strand and Coachella Valley segment of the San Andreas fault zone. We speculate that this is because the 3 km of right slip on the San Bernardino stand of the San Andreas fault and Coachella Valley segment of the Banning fault has not been sufficient to form a single new structure through the 15- to 20-km gap between the two previously unconnected segments. This implies that large earthquake rupture on the San Andreas fault may be inhibited from propagating through San Gorgonio Pass, thus limiting the maximum magnitudes on the southern San Andreas fault. ¿ American Geophysical Union 1996