Regression analysis to develop empirical ground-motion relations that predict ground-motion characteristics as a function of magnitude and distance are an essential part of seismic design and probabilistic seismic hazard assessment. Several different ground-motion relations have been presented based on different data and assumptions. This study evaluates five that were judged likely to be appropriate for southern California. We test them against the strong-motion data recorded in this region between 1933 and 1994, for magnitudes between 5.0 and 7.5, and for distances up to 150 km. The ground-motion parameters examined are peak acceleration and response spectral acceleration, SA, at 0.3-, 1.0-, and 3.0-sec periods. Some of the ground-motion relations are more consistent than others with the limited set of southern California data, but none of them stand out as either “best” or “unacceptable”. Residuals from all of these models indicate that as the magnitude increases the standard deviation decreases at high frequencies but increases for SA (3.0 sec). Among those found to be most consistent, significant differences remain with respect to predicting ground motion under conditions not yet sampled by the existing observational data set (e.g., near-source shaking for large earthquakes). Theoretical simulations are used elsewhere in this report (Ni et al., 2000; Anderson, 2000) to help resolve these ambiguities. The implications with respect to probabilistic seismic hazard assessment are explored in Field and Petersen (2000).