A subset of 3371 events recorded in the Northridge area by the Southern California Seismic Network during January to April 1994 was relocated with the joint hypocentral determination (JHD) technique. This analysis showed two unexpected results: (a) the JHD locations are shifted about 3.9 km on average in a northwest direction with respect to the locations determined using a single-event location (SEL) program, and (b) the station corrections vary between −0.55 and 1.26 sec, a rather large range. In addition, the JHD locations are less scattered than the SEL locations. For each station, the weighted average of the arrival time residuals obtained when the events are located with the SEL program (which does not apply distance or error weighting) are generally smaller than the corresponding JHD corrections. The locations determined with SEL and using the weighted average residuals as station corrections do not differ much from the SEL locations, but on average the RMS residuals become as small as those corresponding to the JHD locations. As the magnitude of the station corrections indicates the presence of large lateral velocity variations, a 3D velocity model for the area was determined using the arrival times of 1012 events recorded by at least 17 stations. The initial velocity model was that used routinely by the Southern California Earthquake Center. The first two layers (5.5- and 10.5-km thick) were subdivided into 100 blocks each (12 × 12 km). These layers show a pronounced low-velocity anomaly (24% and 16%, respectively) immediately to the northwest of the epicentral area. This low-velocity zone coincides with the west Ventura Basin. Another pronounced low-velocity zone to the southeast of the epicentral area reflects the presence of the Los Angeles Basin. The locations obtained with the 3D velocity model are consistently to the southeast of the JHD locations, 2.4 km on average. To establish the effect of these pronounced lateral velocity variations on the SEL and JHD locations, synthetic travel times were analyzed. The synthetic times were generated for event locations determined by JHD (shifted by various amounts) and the 3D velocity model and were subsequently treated as the actual data. The most important result of this analysis is that the JHD locations are affected by a quasi-systematic shift in a northwest direction (up to about 2.7 km on average, depending on the initial shift) but that the relative locations are well preserved. Therefore, both the velocity inversion of the actual data and the analysis of the synthetic data indicate that the JHD locations determined for the actual data are quasi-systematically mislocated. To account for this mislocation, an overall shift of 2.5 km to the southeast was applied to all the JHD locations. One of the most important implications of the shifted locations is the possibility that the northeasterly dipping Santa Susana fault, to the northwest of the epicentral area, was seismically active during the aftershock sequence. This feature is more diffuse in other published locations.