We performed 3D ground-motion simulations for 10 recent small to moderate earthquakes (Mw 4.1–5.4) in the San Francisco Bay area to evaluate two versions of the USGS3D velocity model (Brocher, 2005; Jachens et al., 2006; Brocher, 2008). Comparisons were made in terms of modeling phase arrival timing, peak ground-motion amplitudes, and the seismic waveforms. In the simulations we assumed the source parameters reported in the Berkeley Seismological Laboratory (BSL) Moment Tensor Catalog. Broadband seismic data from the Berkeley Digital Seismic Network (BDSN), and strong motion data from the USGS and the California Geological Survey (CGS) strong motion arrays were used in the analysis. The comparison of peak ground velocity (PGV) for both models reveals that both 3D models predict the observed PGV well over four orders of magnitude, and P- and S-wave timing and pseudospectral acceleration (PSA) are well modeled by the 3D structure. While the revised model (model 8.3.0) significantly improved the timing of the first arrival, and the waveform fit is generally good, there remain discrepancies in estimated amplitudes and durations that require improvements to the structure. Nevertheless, from our low-frequency (0.5 Hz) analysis we found that the 3D model is suitable for the simulation of PGV to assess the strong shaking hazard of future large earthquakes, because earthquakes larger than M 6 have PGV carried by waves of 1 to several seconds period.