There is a growing literature investigating the effects of selenium on the central nervous system and cognitive function. However, little is known about the role of selenoprotein P, the main selenium transporter, which can also have adverse biological effects. We conducted a prospective cohort study of individuals aged 42–81 years who received a clinical diagnosis of mild cognitive impairment. Using sandwich ELISA methods, we measured full-length selenoprotein P concentrations in serum and cerebrospinal fluid to assess the relation with dementia incidence during a median follow-up of 47.3 months. We used Cox proportional hazards regression and restricted cubic splines to model such relation. Of the 54 participants, 35 developed dementia during follow-up (including 26 cases of Alzheimer’s dementia). Selenoprotein P concentrations in serum and cerebrospinal fluid were highly correlated, and in spline regression analyses they each showed a positive non-linear association with dementia risk, particularly after excluding dementia cases diagnosed within 24 months of follow-up. We also observed differences in association according to the dementia subtypes considered. Risk ratios of dementia peaked at 2–6 at the highest levels of selenoprotein P, when compared to its median level, also depending on matrix, analytical methodology and dementia subtype. Findings of this study, the first to assess selenoprotein P levels in the central nervous system in vivo and the first to use a prospective study design to evaluate associations with dementia, suggest that higher circulating concentrations of selenoprotein P, both in serum and cerebrospinal fluid, predict progression of MCI to dementia. However, further confirmation of these findings is required, given the limited statistical precision of the associations and the potential for residual confounding.