BACKGROUND: Observational studies link moderate alcohol consumption to reduced risk of cardiometabolic diseases, including coronary heart disease (CHD) and type 2 diabetes mellitus (T2D). Mendelian randomization (MR) studies suggest that these associations are due to confounding. We present observed and genetically proxied associations between alcohol consumption and the incidence of CHD and T2D among African Americans (AA), European Americans (EA), and Hispanic Americans (HA) from the Million Veteran Program. METHODS: We conducted two retrospective, nested case-control studies of 33,053 CHD and 28,278 T2D cases matched to five controls each at the time of the event (index date). We used the Alcohol Use Disorders Identification Test-Consumption (AUDIT-C) score closest in time prior to the index date to estimate alcohol exposure. Models were adjusted for smoking, body mass index (BMI), chronic kidney disease, rheumatoid arthritis, and the use of statins or antihypertensive medications. MR analyses used either a single variant in ADH1B or a genetic score (GS) as instrumental variables. RESULTS: Observational analysis showed a U-shaped association of alcohol consumption with CHD and T2D risk. However, in MR analyses, neither ADH1B genotype-predicted (in 36,465 AAs, 146,464 EAs, and 11,342 HAs) nor GS-predicted (in EAs) alcohol consumption was associated with CHD risk. Similarly, T2D was not associated with alcohol consumption predicted either by ADH1B genotype (in 42,008 AAs, 109,351 EAs, and 13,538 HAs) or GS (in EAs). Multivariable MR analyses that adjusted for the effects of blood pressure and smoking also showed no association between alcohol consumption and cardiometabolic diseases. CONCLUSIONS: We replicate prior observational studies that show a U-shaped association between alcohol consumption and cardiometabolic diseases, but MR findings show no causal association between these traits. This is largely consistent with previous MR analyses in EAs and expands the literature by providing similar findings in AA and HA populations.