Aims: The association of plasma high-density lipoprotein (HDL) cholesterol with risk of dementia is unclear. We, therefore, tested the hypothesis that high levels of plasma HDL cholesterol are associated with increased risk of dementia and whether a potential association is of a causal nature. Methods and results: In two prospective population-based studies, the Copenhagen General Population Study and the Copenhagen City Heart Study (N = 111 984 individuals), we first tested whether high plasma HDL cholesterol is associated with increased risk of any dementia and its subtypes. These analyses in men and women separately were adjusted multifactorially for other risk factors including apolipoprotein E (APOE) genotype. Second, taking advantage of two-sample Mendelian randomization, we tested whether genetically elevated HDL cholesterol was causally associated with Alzheimer's disease using publicly available consortia data on 643 836 individuals. Observationally, multifactorially adjusted Cox regression restricted cubic spline models showed that both men and women with extreme high HDL cholesterol concentrations had increased risk of any dementia and of Alzheimer's disease. Men in the 96th-99th and 100th vs. the 41st-60th percentiles of HDL cholesterol had multifactorially including APOE genotype adjusted hazard ratios of 1.66 (95% confidence interval 1.30-2.11) and 2.00 (1.35-2.98) for any dementia and 1.59 (1.16-2.20) and 1.87 (1.11-3.16) for Alzheimer's disease. Corresponding estimates for women were 0.94 (0.74-1.18) and 1.45 (1.03-2.05) for any dementia and 0.94 (0.70-1.26) and 1.69 (1.13-2.53) for Alzheimer's disease. Genetically, the two-sample Mendelian randomization odds ratio for Alzheimer's disease per 1 SD increase in HDL cholesterol was 0.92 (0.74-1.10) in the IGAP2019 consortium and 0.98 (0.95-1.00) in the ADSP/IGAP/PGC-ALZ/UKB consortium. Similar estimates were observed in sex stratified analyses. Conclusion: High plasma HDL cholesterol was observationally associated with increased risk of any dementia and Alzheimer's disease, suggesting that HDL cholesterol can be used as an easily accessible plasma biomarker for individual risk assessment.