Higher coffee consumption has been associated with reduced dementia risk, yet with inconsistencies across studies. CYP1A2 polymorphisms, which affects caffeine metabolism, may modulate the association between coffee and the risk of dementia and Alzheimer's disease (AD). We included 5964 participants of the Three-City Study (mean age 74 years-old), free of dementia at baseline when they reported their daily coffee consumption, with available genome-wide genotyping and followed for dementia over a median of 9.0 (range 0.8-18.7) years. In Cox proportional-hazards models, the relationship between coffee consumption and dementia risk was modified by CYP1A2 polymorphism at rs762551 (p for interaction = 0.034). In multivariable-adjusted models, coffee intake was linearly associated with a decreased risk of dementia among carriers of the C allele only ("slower caffeine metabolizers"; HR for 1-cup increased [95% CI] 0.90 [0.83-0.97]), while in non-carriers ("faster caffeine metabolizers"), there was no significant association but a J-shaped trend toward a decrease in dementia risk up to 3 cups/day and increased risk beyond. Thus, compared to null intake, drinking ≥ 4 cups of coffee daily was associated with a reduced dementia risk in slower but not faster metabolizers (HR [95% CI] for ≥ 4 vs. 0 cup/day = 0.45 [0.25-0.80] and 1.32 [0.89-1.96], respectively). Results were similar when studying AD and another CYP1A2 candidate polymorphism (rs2472304), but no interaction was found with CYP1A2 rs2472297 or rs2470893. In this cohort, a linear association of coffee intake to lower dementia risk was apparent only among carriers of CYP1A2 polymorphisms predisposing to slower caffeine metabolism.