© 2017 Japanese College of Cardiology Background Whether smoking increases the risk of atrial fibrillation (AF) remains debatable due to inconsistent reports. Methods We examined the association between smoking and incident AF in 11,047 participants from the REasons for Geographic And Racial Differences in Stroke (REGARDS) Study, one of the largest biracial, population-based cohort studies in the USA. Baseline (2003–2007) cigarette smoking status and amount (pack-years) were self-reported. Incident AF was determined by electrocardiography and history of a prior physician diagnosis at a follow-up examination conducted after a median of 10.6 years. Results During follow-up, 954 incident AF cases were identified; 9.5% in smokers vs. 7.8% in non-smokers; p < 0.001. In a model adjusted for socio-demographics, smoking (ever vs. never) was associated with a 15% increased risk of AF [OR (95%CI): 1.15(1.00, 1.31)], but this association was no longer significant after further adjustment for cardiovascular risk factors [OR (95% CI): 1.12 (0.97, 1.29)]. However, heterogeneities in the association were observed among subgroups; the association was stronger in young vs. old participants [OR (95%CI): 1.31 (1.03, 1.67) vs. 0.99 (0.83–1.18) respectively; interaction p-value = 0.005] and in those with vs. without prior cardiovascular disease [OR (95%CI): 1.18 (0.90, 1.56) vs. 1.06 (0.90, 1.25) respectively; interaction p-value 0.0307]. Also, the association was significant in blacks but not in whites [OR (95%CI): 1.51 (1.12, 2.05) vs. 0.99 (0.84, 1.16), respectively], but the interaction p-value did not reach statistical significance (interaction p-value = 0.65). Conclusions The association between smoking and AF is possibly mediated by a higher prevalence of cardiovascular risk factors in smokers, but there is marked heterogeneity in the strength of this association among subgroups which may explain the conflicting results in prior studies.