In map view, aftershocks of the 2004 Parkfield earthquake lie along a line that forms a straighter connection between San Andreas fault segments north and south of the Parkfield reach than does the mapped trace of the fault itself. A straightedge laid on a geologic map of Central California reveals a ???50-km-long asymmetric northeastward warp in the Parkfield reach of the fault. The warp tapers gradually as it joins the straight, creeping segment of the San Andreas to the north-west, but bends abruptly across Cholame Valley at its southeast end to join the straight, locked segment that last ruptured in 1857. We speculate that the San Andreas fault surface near Parkfield has been deflected in its upper ???6 km by nonelastic behavior of upper crustal rock units. These units and the fault surface itself are warped during periods between large 1857-type earthquakes by the presence of the 1857-locked segment to the south, which buttresses intermittent coseismic and continuous aseismic slip on the Parkfield reach. Because of nonelastic behavior, the warping is not completely undone when an 1857-type event occurs, and the upper portion of the three-dimensional fault surface is slowly ratcheted into an increasingly prominent bulge. Ultimately, the fault surface probably becomes too deformed for strike-slip motion, and a new, more vertical connection to the Earth's surface takes over, perhaps along the Southwest Fracture Zone. When this happens a wedge of material currently west of the main trace will be stranded on the east side of the new main trace.