Vitamin D has an important immunoregulatory function,1 ,2 and there is suggestive evidence for its role in the aetiology of autoimmune diseases, including rheumatoid arthritis (RA).3 Vitamin D levels are reduced in individuals with severe RA,4 ,5 but it is difficult to determine whether this reduction is cause or effect. The onset of arthritis during winter is associated with greater radiographic progression at 6 months, which would be compatible with a causative role.6 Single nucleotide polymorphisms (SNPs) in four vitamin D metabolism genes (GC, DHCR7/NADSYN1, CYP2R1 and CYP24A1) are associated at genome-wide significance with circulating vitamin D levels.7 ,8 Since vitamin D level is associated with RA severity, an association between those SNPs and RA severity would establish causality. Testing the association of genetic variants in genes that control circulating vitamin D levels provides, therefore, an instrument to obtain an unbiased test of the hypothesis that vitamin D levels are aetiologically linked to RA outcome by controlling unmeasured confounding (Mendelian randomisation9). Hence, we tested SNPs previously associated with vitamin D metabolism,7 ,8 and SNPs in these gene regions showing some evidence of association with RA and type 1 diabetes3 ,10 for association with RA outcome in two prospective cohorts of early RA patients with longitudinal radiographic data available (table 1). Cohort characteristics have been described earlier elsewhere.11 Twenty SNPs mapping to four independent genetic regions were successfully genotyped using Sequenom MassArray technology in 1433 patients with 2164 X-rays from the Norfolk Arthritis Register (NOAR) and in 443 RA patients with 2924 X-rays from the Early Rheumatoid Arthritis Study (ERAS) scored using the Larsen method.11 Longitudinal modelling of the presence of erosions was performed using generalised estimating equation (GEE) model with a logit link for binary outcomes, while the Larsen score was modelled with Generalized Linear Latent And Mixed Modelling (GLLAMM) with discrete random effects to construct a finite mixture of three-component normal distributions for the highly skewed Larsen score outcomes with extra zero scores. Effect sizes for GEE models are given as a change in the log odds of developing erosions, and for GLLAMM as an increase in Larsen score, adjusting for age, disease duration and the square of these covariates (table 2). Although serological status and the shared epitope showed consistent association with radiological outcomes in both cohorts, the vitamin D metabolism SNPs showed inconsistent evidence. Thus, rs3829251 in the NADSYN1 gene was associated with Larsen score in ERAS but was not significantly associated with the presence of erosions in the same cohort, nor with radiographic damage in NOAR. Only DHCR7 (rs4944997 and its perfect proxy rs4944076) showed a consistent effect size and achieved the Bonferroni corrected level of significance after meta-analysis using a fixed effects model (Mantel and Haenszel method). A SNP in low linkage disequilibrium with rs4944997 has previously been associated with low vitamin D levels (rs12785878, r²=0.3).7 The protective effect reported for rs4944997 is therefore not consistent with the hypothesis that low vitamin D levels are associated with increased radiographic damage in RA. As the study presented here was performed in one of the largest sample sizes currently available worldwide, the effect of vitamin D SNPs and consequently vitamin D levels is unlikely to play a major role in the aetiology of RA radiographic damage. The reported lower vitamin D levels in more severe RA are more likely to be consequence than cause. Our data do not support vitamin D supplementation as a direct therapeutic intervention to modify RA outcome, although there may be other benefits of vitamin D supplementation in patients with deficient or insufficient levels.