BACKGROUND: Glycaemic traits are associated with poorer brain health and dementia risk. Recent advances in genetic instruments for specific glycaemic markers enable an in-depth investigation of the likely nature of associations and underlying mechanisms between diabetes-related mechanisms and brain health and dementia. METHODS: We used two-sample Mendelian randomisation (MR) in the UK Biobank (UKB) (maximum N = 357 883 White British, mean age 56.9 years, 54% female) applying inverse-variance weighted MR as our main estimator alongside MR-Egger, weighted median estimator (WME) and Mendelian Randomization Pleiotropy RESidual Sum and Outlier (MR-PRESSO) as sensitivity tests. Instruments were 53 insulin resistance, 109 fasting glucose, 48 fasting insulin and 15 2-h post-load glucose genetic variants with variant-outcome effects estimated adjusting for 10 PCs. We checked core MR assumptions and sought to replicate results in an independent Alzheimer's dementia genome-wide association study (GWAS). RESULTS: In UKB, higher 2-h post-load glucose was associated with a 69% increased Alzheimer's dementia risk (odds ratio 1.69 [95% confidence interval 1.38-2.07]), though this did not replicate in an independent GWAS. Fasting insulin, fasting glucose and postprandial glucose did not influence total brain, hippocampal or white-matter hyperintensity volumes. DISCUSSION: The association between elevated 2-h post-load glucose and increased Alzheimer's risk supports a potential role for postprandial hyperglycaemia in dementia. The lack of associations between fasting or postprandial glucose and hippocampal, total-brain or white matter hyperintensity volumes suggests this risk may operate independently of gross structural atrophy. CONCLUSION: Genetically proxied postprandial hyperglycaemia contributes to increased Alzheimer's risk in mid-life, warranting replication in other populations and ancestries to confirm and clarify underlying mechanisms.