Abstract Aims/hypothesis Metformin use has been associated with reduced incidence of dementia in diabetic individuals in observational studies. However, the causality between the two in the general population is unclear. This study uses Mendelian randomisation (MR) to investigate the causal effect of metformin targets on Alzheimer’s disease and potential causal mechanisms in the brain linking the two. Methods Genetic proxies for the effects of metformin drug targets were identified as variants in the gene for the corresponding target that associated with HbA_1c level (N=344,182) and expression level of the corresponding gene (N≤31,684). The cognitive outcomes were derived from genome-wide association studies comprising 527,138 middle-aged Europeans, including 71,880 with Alzheimer’s disease or Alzheimer’s disease-by-proxy. MR estimates representing lifelong metformin use on Alzheimer’s disease and cognitive function in the general population were generated. Effect of expression level of 22 metformin-related genes in brain cortex (N=6601 donors) on Alzheimer’s disease was further estimated. Results Genetically proxied metformin use, equivalent to a 6.75 mmol/mol (1.09%) reduction on HbA_1c, was associated with 4% lower odds of Alzheimer’s disease (OR 0.96 [95% CI 0.95, 0.98], p=1.06×10⁻⁴) in non-diabetic individuals. One metformin target, mitochondrial complex 1 (MCI), showed a robust effect on Alzheimer’s disease (OR 0.88, p=4.73×10⁻⁴) that was independent of AMP-activated protein kinase. MR of expression in brain cortex tissue showed that decreased MCI-related gene (NDUFA2) expression was associated with lower Alzheimer’s disease risk (OR 0.95, p=4.64×10⁻⁴) and favourable cognitive function. Conclusions/interpretation Metformin use may cause reduced Alzheimer’s disease risk in the general population. Mitochondrial function and the NDUFA2 gene are plausible mechanisms of action in dementia protection. Graphical abstract