The attenuation relationship presented by Boore et al. (1997) has been evaluated and customized with respect to southern California strong-motion data (for peak ground acceleration (PGA) and 0.3-, 1.0-, and 3.0-sec period spectral acceleration). This study was motivated by the recent availability of a new site-classification map by Wills et al. (2000), which distinguishes seven different site categories for California based on the 1994 NEHRP classification. With few exceptions, each of the five site types represented in the southern California strong-motion database exhibit distinct amplification factors, supporting use of the Wills et al. (2000) map for microzonation purposes. Following other studies, a basin-depth term was also found to be significant and therefore added to the relationship. Sites near the center of the LA Basin exhibit shaking levels up to a factor of 2 greater, on average, than otherwise equivalent sites near the edge. Relative to Boore et al. (1997), the other primary difference here is that PGA exhibits less variation among the Wills et al. (2000) site types. In fact, the PGA amplification implied by the basin-depth effect is greater than that implied by site classification. The model does not explicitly account for nonlinear sediment effects, which, if important, will most likely influence rock-site PGA predictions the most. Evidence for a magnitude-dependent variability, or prediction uncertainty, is also found and included as an option.