Objective Studies suggest that circulating type I interferon (IFN) may predict response to biological agents in rheumatoid arthritis (RA). Prediction of response prior to initiating therapy would represent a major advancement. Methods We studied sera from a test set of 32 patients with RA from the Auto-immune Biomarkers Collaborative Network Consortium and a validation set of 92 patients with RA from the Treatment Efficacy and Toxicity in Rheumatoid Arthritis Database and Repository registry. The test set included those with good response or no response to tumour necrosis factor (TNF) inhibitors at 14 weeks by European League Against Rheumatism criteria. The validation set included subjects with good, moderate or no response at 12 weeks. Total serum type I IFN activity, IFN-α and IFN-β activity were measured using a functional reporter cell assay. Results In the test set, an increased ratio of IFN-β to IFN-α (IFN-β/α activity ratio) in pretreatment serum associated with lack of response to TNF inhibition (p=0.013). Anti-cyclic citrullinated peptide antibody titre and class of TNF inhibitor did not influence this relationship. A receiver-operator curve supported a ratio of 1.3 as the optimal cut-off. In the validation set, subjects with an IFN-β/α activity ratio >1.3 were significantly more likely to have non-response than good response (OR=6.67, p=0.018). The test had 77% specificity and 45% sensitivity for prediction of non-response compared with moderate or good response. Meta-analysis of test and validation sets confirmed strong predictive capacity of IFN-β/α activity ratio (p=0.005). Conclusions Increased pretreatment serum IFN-β/α ratio strongly associated with non-response to TNF inhibition. This study supports further investigation of serum type I IFN in predicting outcome of TNF inhibition in RA.