The present study used a systematic review approach to identify relevant randomised control trials (RCT) with vitamin D and then apply meta-regression to explore the most appropriate model of the vitamin D intake–serum 25-hydroxyvitamin D (25(OH)D) relationship to underpin setting reference intake values. Methods included an updated structured search on Ovid MEDLINE; rigorous inclusion/exclusion criteria; data extraction; and meta-regression (using different model constructs). In particular, priority was given to data from winter-based RCT performed at latitudes >49·5°N (n 12). A combined weighted linear model meta-regression analyses of natural log (Ln) total vitamin D intake (i.e. diet and supplemental vitamin D) v. achieved serum 25(OH)D in winter (that used by the North American Dietary Reference Intake Committee) produced a curvilinear relationship (mean (95 % lower CI) serum 25(OH)D (nmol/l) = 9·2 (8·5) Ln (total vitamin D)). Use of non-transformed total vitamin D intake data (maximum 1400 IU/d; 35 μg/d) provided for a more linear relationship (mean serum 25(OH)D (nmol/l) = 0·044 × (total vitamin D)+33·035). Although inputting an intake of 600 IU/d (i.e. the RDA) into the 95 % lower CI curvilinear and linear models predicted a serum 25(OH)D of 54·4 and 55·2 nmol/l, respectively, the total vitamin D intake that would achieve 50 (and 40) nmol/l serum 25(OH)D was 359 (111) and 480 (260) IU/d, respectively. Inclusion of 95 % range in the model to account for inter-individual variability increased the predicted intake of vitamin D needed to maintain serum 25(OH)D ≥ 50 nmol/l to 930 IU/d. The model used to describe the vitamin D intake–status relationship needs to be considered carefully when setting new reference intake values in the Europe.