Seismograms of earthquakes and explosions recorded at local, regional, and teleseismic distances by a small-aperture, dense seismic array located on Pinyon Flat, in southern California, reveal large (±15°) backazimuth anomalies. We investigate the causes and implications of these anomalies by first comparing the effectiveness of estimating backazimuth with an array using three different techniques: the broadband frequency-wavenumber (BBFK) technique, the polarization technique, and the beamforming technique. While each technique provided nearly the same direction as a most likely estimate, the beamforming estimate was associated with the smallest uncertainties. Backazimuth anomalies were then calculated for the entire data set by comparing the results from beamforming with backazimuths derived from earthquake locations reported by the Anza and Caltech seismic networks and the Preliminary Determination of Epicenters (PDE) Bulletin. These backazimuth anomalies have a simple sinelike dependence on azimuth, with the largest anomalies observed from the southeast and northwest directions. Such a trend may be explained as the effect of one or more interfaces dipping to the northeast beneath the array. A best-fit model of a single interface has a dip and strike of 20° and 315°, respectively, and a velocity contrast of 0.82 km/sec. Application of corrections computed from this simple model to ray directions significantly improves locations at all distances and directions, suggesting that this is an upper crustal feature. We confirm that knowledge of local structure can be very important for earthquake location by an array but also show that corrections computed from simple models may not only be adequate but superior to those determined by raytracing through smoothed laterally varying models.