A small aperture (≈300 m), four-station array was deployed in Sunnyvale, California for 5 days to record aftershocks of the Loma Prieta earthquake of October 1989. The purpose of the array was to study the seismic response of the alluvium-filled Santa Clara Valley and the role of surface waves in the seismic shaking of sedimentary basins. Strong-motion records of the Loma Prieta mainshock indicate that surface waves produced the peak velocities and displacements at some sites in the Santa Clara Valley. We use the recordings from the dense array to determine the apparent velocity and azimuth of propagation for various arrivals in the seismograms of four aftershocks with magnitudes between 3.6 and 4.4. Apparent velocities are generally observed to decrease with increasing time after the S wave in the seismograms. Phases arriving less than about 8 sec after the S wave have apparent velocities comparable to the S wave and appear to be body waves multiply reflected under the receiver site or reflected by crustal interfaces. For times 10 to 30 sec after the direct S wave, we observe long-period (1 to 6 sec) arrivals with apparent velocities decreasing from 2.5 to 0.8 km / sec. We interpret these arrivals to be surface waves and conclude that these surface waves produce the long duration of shaking observed on the aftershock records. Much of the energy in the 40 sec after the S-wave is coming approximately from the direction of the source, although some arrivals have backazimuths as much as 60° different from the backazimuths to the epicenters. Two of the aftershocks show arrivals coming from 30 to 40° more easterly than the epicenters. This energy may have been scattered from outcrops along the southeastern edge of the basin. In contrast, the deepest aftershock studied (d = 17 km) displays later arrivals with backazimuths 30 to 40° more westerly than the epicenter. A distinct arrival for one of the aftershocks propagates from the southwest, possibly scattered from the western edge of the basin. Synthetic seismograms derived from a plane-layered crustal model do not produce the long-period Love waves observed in the waveforms of the ML 4.4 aftershock. These Love waves may be generated by the conversion of incident S waves or Rayleigh waves near the edge of the basin.