3D velocity models constitute a key element in strong ground motion modeling, for example, earthquake-hazard assessment. Their validation is typically based on modeling weak earthquakes with foci limited to depths greater than ~5 km. However, ruptures during moderate and large earthquakes can propagate to shallower depths (and eventually reach the surface). For such shallow sources, velocity models may not be validated with sufficient accuracy. In this respect, we conduct a series of tests based on the 2014 Mw 6.0 South Napa earthquake, which was characterized by a very shallow slip asperity, to assess the performance of the U.S. Geological Survey 3D San Francisco Bay area velocity model within 20-km fault distance. Our study indicates that the velocity model performs generally well with some exceptions, in which large-amplitude surface waves not present in the observed data are systematically excited. We conclude that more complex fault geometries or slightly deeper slip would not result in a better fit of the observed data. Contrarily, we demonstrate that smoothing the velocity model (i.e., reducing the strong velocity contrasts between basin fill and bedrock) effectively attenuates the spurious oscillations.