Objective To investigate the associations between low and high concentrations of baseline serum 25-hydroxyvitamin D [25(OH)D] and all-cause mortality in very old (≥85 years) men and women over 6 years. Design, setting and subjects Prospective mortality data from 775 participants in the Newcastle 85+ Study were analysed for survival in relation to 25(OH)D (season-specific quartiles and predefined cut-off values) and sex using Cox proportional hazards models. The models were fitted to the entire and restricted (nonusers of vitamin D-containing supplements and medication) cohorts. Results For the entire cohort, mortality was higher in both the lowest and highest 25(OH)D season-specific quartiles [SQ1: hazard ratio (HR) 1.31, 95% confidence interval (CI) 1.01–1.69, P = 0.04; SQ4: HR 1.44, 95% CI 1.12–1.85, P = 0.004] compared with the combined middle quartiles (SQ2 + SQ3), after adjustment for sociodemographic factors. The increased risk for the highest quartile remained significant after further adjustment for lifestyle variables (SQ4: HR 1.37, 95% CI 1.06–1.77, P = 0.02) and was seen only in women in sex-specific analyses. Similarly, in sensitivity analyses with predefined 25(OH)D cut-off values, the highest 25(OH)D concentration (≥75 nmol L−1) was associated with a 2.4-fold increased risk of mortality in women (restricted cohort) after adjusting for all covariates. Conclusion Low and high season-specific 25(OH)D quartiles were associated with increased risks of mortality over 6 years in the very old; this effect was particularly noticeable in women, including those who reported taking vitamin D-containing supplements/medication.