The number of deliveries by cesarean section (c-section) has increased dramatically. Clinical and demographic factors have not adequately explained the increased rate, however. This study investigates the role of nonclinical (i.e., physician) factors in explaining variations in c-section rates, including the physician's training/experience, financial and convenience incentives, and practice characteristics. The study measures the impact of these factors on the decision to perform a c-section rather than opting for vaginal delivery, controlling for a host of patient and hospital characteristics. Physician effects are evaluated in terms of their overall contribution to the explanatory power of logistic regression models, as well as in terms of specific hypotheses to be tested. The analyses are based on 33,233 deliveries performed by 441 physicians in 36 hospitals in 1 state during 1989. As a set, physician factors contribute more to the explanatory power of the model than do hospital factors, despite being added last to the equation. Parameter estimates provide more support for the hypothesized effects of physician convenience incentives than background/training. The log odds of performing a c-section increase with the physician's rate of c-sections in the prior year, delivery on a Friday, and delivery between 6 AM and 6 PM, and decrease with the concentration of the physician's hospital practice. Patient factors appear much more important than both physician and hospital factors, however. Efforts to reduce unnecessary c-sections should focus on identifying the appropriate clinical indications for c-section and disseminating this information to physicians.