We studied the association between mid-life smoking and late-life dementia in the Honolulu Heart Program (1965-1971) and follow-up assessment for dementia (1991-1996) of 3734 Japanese-American men (80% of survivors). Neuropathologic data were available for 218 men. Adjusting for age, education and apolipoprotein E (APOE) genotype, the risk of Alzheimer's disease (AD) in smokers increased with pack-years of smoking at medium (odds ratio (OR)=2.18, 95% confidence interval (CI)=1.07-4.69) and heavy (OR=2.40; 95% CI=1.16-5.17) smoking levels. Very heavy smoking was not associated with AD (OR=1.08; 95% CI=0.43-2.63). Findings were similar when AD cases included those with cerebrovascular disease and for all dementias combined. Adjustment for cardiovascular and respiratory factors or stratification by apolipoprotein E genotype did not change these associations. In an autopsied subsample, the number of neuritic plaques increased with amount smoked. This study suggests that amount smoked is associated with an increasing risk of AD and Alzheimer-type neuropathology up to heavy smoking levels. The lack of association in very heavy smokers may be due to a hardy survivor effect.