We examined CYP1A1 T6235C (M1) and A4889G (M2) polymorphisms in San Francisco Bay Area African Americans and Latinos who were newly diagnosed with primary lung cancer from September 1998 to November 2002 and in age-gender-ethnicity frequency-matched controls. Owing mainly to rapid mortality of cases, overall percentages of cases genotyped were 26% and 32% for Latinos and African Americans, respectively. CYP1A1 variants were genotyped for Latinos (104 cases, 278 controls) and African Americans (226 cases, 551 controls). M1 and M2 frequencies in controls were 0.23 and 0.02 for African Americans and 0.38 and 0.29 for Latinos. In Latinos, the overall inverse odds ratio (OR) of 0.51 (95% CI = 0.32-0.81) for M1 variant genotype resulted from an inverse interaction with smoking. Nonsmokers with M1 genotype had a slight elevated OR (1.5; 0.59-3.7), but those with less than 30 or 30 or more pack-year history had 0.20 (0.06-0.70) and 0.21 (0.06-0.81) times (about 1/5) the odds expected if smoking and genotype were independent lung cancer risk factors. African Americans had interactions of similar magnitude that were not statistically significant. Results for M2 were very similar. Inverse interactions of CYP1A1 variants and smoking-associated lung cancer risk in Latinos might be causal, due to undetected bias or confounding, or represent a unique linkage disequilibrium between a new lung cancer locus and CYP1A1 in this highly admixed population.