Phosphodiesterase-5 (PDE5) inhibitors have gained interest as a potential treatment for dementia. However, current evidence is limited to observational and pre-clinical studies. We conducted a drug-target Mendelian randomization (MR) analysis to investigate the on-target effects of pharmacological PDE5 inhibition on dementia subtypes and related phenotypes. We selected variants from around the PDE5A locus associated with diastolic and systolic blood pressure, as well as circulating PDE5A levels, to create three instruments for genetically proxied PDE5A inhibition. Using two-sample MR, we validated the instruments against erectile dysfunction and pulmonary arterial hypertension before assessing their associations with dementia subtypes, dementia-related proteins, and neuroimaging traits. After correcting for multiple comparisons, genetically proxied PDE5 inhibition, per one SD lower in diastolic blood pressure, was associated with higher odds of Alzheimer's disease (OR 1.09, 95% CI 1.07-1.11) and Lewy body dementia (OR 1.32, 95% CI 1.23-1.41), but a trend towards lower odds of vascular dementia across all instruments. Genetically proxied PDE5 inhibition was associated with both beneficial and adverse effects on brain MRI traits. This included lower volumes of white matter hyperintensities (SD change - 0.035, 95% CI - 0.025, - 0.045), indicating potential benefits, but also reduced volumes of other structures, including the thalamus, suggesting potential adverse effects. PDE5 inhibition was associated with the concentrations of several proteins implicated in dementia pathophysiology. Our findings suggest that while PDE5 inhibition may be associated with a lower risk of vascular dementia, possibly by preventing white matter hyperintensities, it may increase risk of Alzheimer's disease and Lewy body dementia, warranting further investigation before clinical trials.