Using a waveform cross-correlation technique, Rubin and Gillard <2000> obtained precise relative locations for 4300 0.5 < M < 3.5 earthquakes occurring along 50 km of the San Andreas fault. This study adds to that another 5000 earthquakes distributed along 10 km of the San Andreas fault and 20 km of the Calaveras fault. Errors in relative location are typically tens of meters for earthquakes separated by hundreds of meters and, after correcting for time-dependent station delays, meters for earthquakes separated by tens of meters. Along both faults, the minimum separation between consecutive earthquakes scales with magnitude in a manner consistent with a magnitude-independent stress drop. By treating each earthquake on the San Andreas as if it were a main shock, scaling the distances to all subsequent earthquakes by main shock size, and stacking the results, a composite aftershock sequence is produced that has many of the characteristics predicted by rate-and-state friction models. Projected onto the fault surface, these aftershocks outline a quasi-elliptical, roughly 4-MPa stress drop main shock elongate in the slip-parallel direction by ~40%. At the ends of the major axes of this ellipse over twice as many aftershocks occur to the NW than to the SE, an asymmetry attributed to the contrast in material properties across the fault. Unlike the San Andreas, the Calaveras fault exhibits little P wave velocity contrast and no discernible aftershock asymmetry; however, the earliest part of the aftershock sequence on the Calaveras might be truncated by the ~30-s blind time of the network following a triggering event.