We relocate earthquakes that occurred near the 1988 (ML=4.7) and the 1990 (ML=5.5) Upland, California, earthquakes to map the fault geometry of the poorly defined San Jose fault and to test the static stress triggering hypothesis for this sequence. We adopt the L1 norm, waveform cross-correlation method of Shearer <1997> to obtain precise relocations for 1573 events between 1981 and 1997 in the Upland area. To limit computation time, we only perform waveform cross correlation on 60 of the nearest neighbors of each relocated event. Our final relocations show two linear features. The first is imaged by the locations of the initial month of aftershocks of the 1988 Upland earthquake, which delineate a fault with a dip angle of ~45¿ between 7 and 9 km depth, consistent with the mainshock focal mechanism. The second linear feature is a plane dipping at about 74¿ from 2 to 9 km depth, which is illuminated by both the 1988 and 1990 Upland sequences, in agreement with the inferred location of the San Jose fault at depth. However, below 9 km the event locations become more diffuse, giving rise to two different interpretations of the fate of the San Jose fault at depth. One possibility is that the fault shallows at depth, consistent with our relocations but not with the focal mechanism of a ML=4.7 deep aftershock. Alternatively, the fault may be offset at depth by the more shallow dipping fault strand broken during the 1988 earthquake. Using these inferred fault geometries, we compute stress changes resulting from slip during the mainshocks to test whether the relocated aftershocks are consistent with the hypothesis that more aftershocks occur where the change in static Coulomb failure stress is positive (on faults optimally oriented for failure). This requires an extension of previous models of changes in the failure stress to three dimensions and arbitrary fault orientation. We find that patterns of change in Coulomb failure stress differ little between the different fault geometries: all are nearly symmetric about the fault and so do not match the aftershock distribution, in which most of the off-fault events occur to one side of the fault plane. ¿ 2000 American Geophysical Union