BackgroundSelenium has potential safeguarding properties against cognitive decline, because of its role in protecting DNA, proteins, and lipids in the brain from oxidative damage. However, acute and chronic overexposure to selenium can be neurotoxic. ObjectiveThe aim of this analysis was to explore the association between selenium status [serum selenium and selenoprotein P (SELENOP) concentrations and glutathione peroxidase 3 (GPx3) activity] and cognitive function in 85-y olds living in Northeast England at baseline and ≤5 y of follow-up. MethodsGlobal cognitive performance was assessed in 755 participants from the Newcastle 85+ study using the standardized Mini-Mental State Examination and attention-specific cognition was assessed using composite scores derived from the Cognitive Drug Research System. Serum selenium, SELENOP, and GPx3 activity were measured at baseline by total reflection X-ray fluorescence, enzyme-linked immunosorbent assay, and coupled-enzyme reaction, respectively. Regression analyses explored linear and nonlinear associations between continuous values and tertiles of selenium status biomarkers, respectively, and cognitive function at baseline. Generalized linear mixed models explored associations between continuous values and tertiles of selenium status biomarkers, and global cognitive decline over 5 y, and attention-specific cognitive decline over 3 y. ResultsOver 3 and 5 y, none of the selenium biomarkers were associated with the rate of cognitive decline. At baseline, in fully adjusted models, higher serum selenium was nonlinearly associated with global cognition (β = 0.05 ± 0.01, P = 0.387 linear, β = 0.04 ± 0.01, P = 0.002 nonlinear). SELENOP and GPx3 activity were not associated with any cognitive outcomes. ConclusionsThere were no associations between selenium status and cognitive decline. However, serum selenium, but not SELENOP or GPx3 activity, was positively associated nonlinearly with global cognition at baseline. Furthermore, these associations were not evident during follow-up, potentially because of residual confounding and reverse causation.