This population-based study examines the factors affecting referrals by primary care physicians (PCPs) to specialists. Multilevel Poisson models were used to test the impact of patient, physician and community-level variables on the referral rate (the number of office-based specialist referrals per patient by the patient's customary PCP in fiscal year 1997/98). Patients from each of 6972 PCPs with sufficient data in Ontario were examined. The average patient had 0.56 referrals per year (range 0-61). Referrals were higher at ages 1 and 77 to 78, and among women of childbearing age. Chronic disease variables were strongly correlated with referral rates. Patients in poor neighborhoods had more referrals, because they had more chronic diseases. After controlling for disease, individuals in the top 9% wealthiest neighborhoods had 4% more referrals. Female physicians made 8% more referrals than men. Older physicians referred more because they saw older patients; after controlling for patient age, physician age had no effect. Referrals were 14% higher in cities with medical schools compared with other cities and 12% lower in small towns. However, local specialist supply was unrelated to referral rates. This study improves our understanding of the impact of physician gender and age on referrals. It suggests that community type, not specialist supply, predicts variations in referrals. Lastly, it identifies preferential access to specialists among high-income earners, even within Canada's universal health insurance system. However, this effect is modest, suggesting that the system does provide reasonably equitable access to referrals.