This study determines site-response factors that can be applied as corrections to a rock-attenuation relationship for use in probabilistic seismic-hazard analysis. The site-response factors are amplitude and site-class dependent. These amplification factors are determined by averaging ratios between observed and predicted ground motions for peak ground acceleration (PGA) and for 5% damped response spectral acceleration at 0.3, 1.0, and 3.0 sec oscillator periods. The observations come from the strong-motion database of the Southern California Earthquake Center (SCEC), and the predictions are based on the Sadigh (1993) rock-attenuation relation. When separated and averaged according to surface geology, significantly different site-response factors are found for Quaternary and Mesozoic units, but a subclassification of Quaternary is generally not justified by the data. The low inputmotion amplification factors are consistent with those obtained from independent aftershock studies at the PGA and 0.3-second period. An observed trend of decreasing Quaternary site amplification with higher input motion is consistent with nonlinear soil behavior; however, the trend exists for Mesozoic sites as well, implying that this may be an artifact of the Sadigh relationship. There is a correlation between larger site-response factors and lower average shear-wave velocity in the upper 30 m for low predicted PGA input motions, with an increase in the correlation with increasing period. The 0.3-sec site response factors for Quaternary data in southern California determined in this study are consistent with 0.3-sec NEHRP site-response correction factors; however, at 1.0-sec period some inconsistencies remain. A trend is also seen with respect to sediment basin depth, where deeper sites have higher average site-response factors. These results constitute a customized attenuation relationship for southern California. The implication of these customized attenuation models with respect to probabilistic hazard analysis is examined in Field and Petersen (2000).