Background: Observational evidence suggests that higher educational attainment is protective for Alzheimer's disease (AD). It is unclear whether this association is causal or confounded by demographic and socioeconomic characteristics. We examined the causal effect of educational attainment on AD in a two-sample MR framework. Methods: We extracted all available effect estimates of the 74 single nucleotide polymorphisms (SNPs) associated with years of schooling from the largest genome-wide association study (GWAS) of educational attainment (N=293,723) and the GWAS of AD conducted by the International Genomics of Alzheimer's Project (n=17,008 AD cases and 37,154 controls). SNP-exposure and SNP-outcome coefficients were combined using an inverse variance weighted approach, providing an estimate of the causal effect of each SD increase in years of schooling on AD. We also performed appropriate sensitivity analyses examining the robustness of causal effect estimates to the various assumptions and conducted simulation analyses to examine potential survival bias of MR analyses. Findings: With each SD increase in years of schooling (3.51 years), the odds of AD were, on average, reduced by approximately one third (odds ratio=0.63, 95% confidence interval [CI]: 0.48 to 0.83, p<0.001). Causal effect estimates were consistent when using causal methods with varying MR assumptions or different sets of SNPs for educational attainment, lending confidence to the magnitude and direction of effect in our main findings. There was also no evidence of survival bias in our study. Interpretation: Our findings support a causal role of educational attainment on AD, whereby an additional ~3.5 years of schooling reduces the odds of AD by approximately one third.