OBJECTIVE: To examine primary care physician referral rate variations, including their extent and their stability over time and across diagnostic categories. DATA SOURCES: 1995/1996 claims data for adult patients from a large Independent Practitioner Association (IPA) model managed care organization (MCO) in the Rochester, NY metropolitan area. The IPA includes over 95 percent of area primary care physicians (PCPs), and the MCO includes over 50 percent area residents. STUDY DESIGN: Referral rates (patients referred to and seen by specialists one or more times/patients seen by PCP/year) were developed for the PCPs (457 general practitioners, family physicians, and internists) in the MCO, including observed referral rates, expected referral rates based on case-mix adjustment across the whole sample, physician-specific case mix-adjusted referral rates (empirical Bayes estimates), and diagnostic category-specific case mix-adjusted referral rates. PRINCIPAL FINDINGS: Wide variations in observed referral rates (0.01-0.69 patients referred/patients seen/year) were attenuated relatively little by case-mix adjustment and persisted in case mix-adjusted empirical Bayes estimates (0.02-0.65). The year-to-year case mix-adjusted referral rate correlation was .90. Correlations of case mix adjusted-referral rates across diagnostic categories were moderate (r=.46-.67). CONCLUSIONS: PCP referral rates exhibit wide variations that are independent of case mix, remain stable over time, and are generalizable across diagnostic categories. Understanding this physician practice variation and its relationship to costs and outcomes is critical to evaluating the effect of current efforts to reduce PCP referral rates.