OBJECTIVES: To estimate the exposure-response function associating polycyclic aromatic hydrocarbon (PAH) exposure and lung cancer, with consideration of smoking. METHODS: Mortality, occupational exposure and smoking histories were ascertained for a cohort of 16,431 persons (15,703 men and 728 women) who had worked in one of four aluminium smelters in Quebec from 1950 to 1999. A variety of exposure-response functions were fitted to the cohort data using generalised relative risk models. RESULTS: In 677 lung cancer cases there was a clear trend of increasing risk with increasing cumulative exposure to PAH measured as benzo(a)pyrene (BaP). A linear model predicted a relative risk of 1.35 (95% CI 1.22 to 1.51) at 100 microg/m(-3) BaP years, but there was a significant departure from linearity in the direction of decreasing slope with increasing exposures. Among the models tried, the best fitting were a two-knot cubic spline and a power curve (RR = (1+bx)(p)), the latter predicting a relative risk of 2.68 at 100 microg/m(-3) BaP years. Additive models and multiplicative models for combining risks from occupational PAH and smoking fitted almost equally well, with a slight advantage to the additive. CONCLUSION: Despite the large cohort with long follow-up, the shape of the exposure-response function and the mode of combination of risks due to occupational PAH and smoking remains uncertain. If a linear exposure-response function is assumed, the estimated slope is broadly in line with the estimate from a previous follow-up of the same cohort, and somewhat higher than the average found in a recent meta-analysis of lung cancer studies.