Many methods for estimating site response compare ground motions at sites of interest to a nearby rock site that is considered a “reference” motion. The critical assumption in these methods is that the surface-rock-site record (reference) is equivalent to the input motion at the base of the soil layers. Data collected in this study show that surface-rock sites can have a site response of their own, which could lead to an underestimation of the seismic hazard when these sites are used as reference sites. Data were collected from local and regional earthquakes on digital recorders, both at the surface and in boreholes, at two rock sites and one basin site in the San Jacinto mountains, southern California. The two rock sites, Keenwild and Piñon Flat, are located on granitic bedrock of the southern California peninsular ranges batholith. The basin site, Garner Valley, is an ancestral lake bed with watersaturated sediments, on top of a section of decomposed granite, which overlies the competent bedrock. Ground motion is recorded simultaneously at the surface and in the bedrock at all three sites. When the surface-rock sites are used as the reference site, i.e., the surface-rock motion is used as the input to the basin, the computed amplification underestimates the actual amplification at the basin site for frequencies above 2 to 5 Hz. This underestimation, by a factor of 2 to 4 depending on frequency and site, results from the rock sites having a site response of their own above the 2-to 5-Hz frequencies. The near-surface weathering and cracking of the bedrock affects the recorded ground motions at frequencies of engineering interest, even at sites that appear to be located on competent crystalline rock. The bedrock borehole ground motion can be used as the reference motion, but the effect of the downgoing wave field and the resulting destructive interference must be considered. This destructive interference may produce pseudo-resonances in the spectral amplification estimates. If one is careful, the bedrock borehole ground motion can be considered a good reference site for seismic hazard analysis even at distances as large as 20 km from the soil site.