To estimate effects of vitamin D levels on incident delirium hospital admissions using inherited genetic variants in mendelian randomization models, which minimize confounding and exclude reverse causation. Longitudinal analysis using the UK Biobank, community-based, volunteer cohort (2006-2010) with incident hospital-diagnosed delirium (ICD-10 F05) ascertained during ≤9.9 years of follow-up of hospitalization records (to early 2016). We included volunteers of European descent aged 60-plus years by end of follow-up. We used single-nucleotide polymorphisms previously shown to increase circulating vitamin D levels, and APOE variants. Cox competing models accounting for mortality were used. Of 313,121 participants included, 544 were hospitalized with delirium during follow-up. Vitamin D variants were protective for incident delirium: hazard ratio = 0.74 per 10 nmol/L (95% confidence interval 0.62-0.87, p = 0.0004) increase in genetically instrumented vitamin D, with no evidence for pleiotropy (mendelian randomization-Egger p > 0.05). Participants with ≥1 APOE ε4 allele were more likely to develop delirium (e.g., ε4ε4 hazard ratio = 3.73, 95% confidence interval 2.68-5.21, p = 8.0 × 10-15 compared to ε3ε3), but there was no interaction with vitamin D variants. In a large community-based cohort, there is genetic evidence supporting a causal role for vitamin D levels in incident delirium. Trials of correction of low vitamin D levels in the prevention of delirium are needed.