Seismograms from the 1995 to 1996 Ridgecrest, California, earthquake sequence, recorded by the TriNet digital seismic network, provide high-quality waveforms from sites throughout southern California, including sites in markedly heterogeneous areas like the Los Angeles area sedimentary basins. Synthetic seismograms calculated by the reflectivity method with various 1D models are used as a baseline to measure the variability of amplitudes throughout southern California. Regardless of the model used, there is greater variability in the amplitudes from basin site records than from rock site records. Rock, soil, and basin sites are all rather insensitive to radiation pattern nodes at the three frequency bands investigated: 0.1 to 0.2 Hz, 0.2 to 0.4 Hz, and 0.4 to 0.8 Hz. This complicates the analysis because the nodes create singularity points in the distribution of ratios of observed and synthetic amplitudes. When stations near nodal planes are removed, the surface waves observed at most rock sites have peak amplitudes within a factor of 2 of synthetic waveform amplitudes. Peak amplitude of the surface waves observed at the soil and basin stations are more variable, with the bulk of the distribution of data/synthetic amplitude ratios less than 3 and a few outliers greater than 5. These outliers occur at the higher frequency bands. Soil and basin sites are also more often larger than the synthetics (higher median values). Most outliers can be explained by applying a water level of 50% to the radiation pattern. This reduces the scatter in the distributions to about the same extent as removing data within 10° of nodes. Thus, most of the outliers are sites that are insensitive to the nodes, not sites that are larger than the overall data distribution.