The objectives of this study were: (1) to describe the issues in developing a clinical predictor of cesarean delivery that could be used to adjust reported cesarean rates for case mix; and (2) to compare its performance to other, simpler predictors using clinical and statistical criteria. Data sources were singleton births greater than 2,500 grams in Washington State in 1989 and 1990 for whom mothers and infant hospital discharge records could be matched to birth certificate data. The study found that merged data led to better predictor variables than those based on one source. A simple four-category hierarchical classification into births with prior cesarean, breech but no prior cesarean, first birth, and other explains 30 percent of the variance in individual cesarean rates. The full clinical model fit the data well and explained 37 percent of the variance. Multiparas without serious complications comprised 35 percent of the mothers and averaged less than 2 percent cesareans. A hospital's predicted cesarean rate depends strongly on the proportion of its births that are first births.