Higher milk intake has been associated with a lower stroke risk, but not with risk of CHD. Residual confounding or reverse causation cannot be excluded. Therefore, we estimated the causal association of milk consumption with stroke and CHD risk through instrumental variable (IV) and gene-outcome analyses. IV analysis included 29 328 participants (4611 stroke; 9828 CHD) of the European Prospective Investigation into Cancer and Nutrition (EPIC)-CVD (eight European countries) and European Prospective Investigation into Cancer and Nutrition-Netherlands (EPIC-NL) case-cohort studies. rs4988235, a lactase persistence (LP) SNP which enables digestion of lactose in adulthood was used as genetic instrument. Intake of milk was first regressed on rs4988235 in a linear regression model. Next, associations of genetically predicted milk consumption with stroke and CHD were estimated using Prentice-weighted Cox regression. Gene-outcome analysis included 777 024 participants (50 804 cases) from MEGASTROKE (including EPIC-CVD), UK Biobank and EPIC-NL for stroke, and 483 966 participants (61 612 cases) from CARDIoGRAM, UK Biobank, EPIC-CVD and EPIC-NL for CHD. In IV analyses, each additional LP allele was associated with a higher intake of milk in EPIC-CVD (β = 13·7 g/d; 95 % CI 8·4, 19·1) and EPIC-NL (36·8 g/d; 95 % CI 20·0, 53·5). Genetically predicted milk intake was not associated with stroke (HR per 25 g/d 1·05; 95 % CI 0·94, 1·16) or CHD (1·02; 95 % CI 0·96, 1·08). In gene-outcome analyses, there was no association of rs4988235 with risk of stroke (OR 1·02; 95 % CI 0·99, 1·05) or CHD (OR 0·99; 95 % CI 0·95, 1·03). Current Mendelian randomisation analysis does not provide evidence for a causal inverse relationship between milk consumption and stroke or CHD risk.