The rapidity with which bone and cartilage damage occurs in patients with rheumatoid arthritis (RA), and the increasing body of evidence for the effectiveness of early intervention in RA, mean that there is a great need for approaches to accurately predict the development of RA in patients with early undifferentiated arthritis. We will review developments in the prediction of outcome on the basis of clinical and laboratory features, including measures of anti-citrullinated protein/peptide antibody status. Although accurate predictions are possible in the majority of patients using recently developed predictive algorithms which utilize clinical and serological variables, there remains a group of patients for whom it is very difficult to predict the development of RA. The utility of new strategies for prediction will be discussed, including recently discovered genetic associations of RA, an assessment of material from the primary site of pathology (the joint), and assessment using the highly sensitive imaging modalities of ultrasound and magnetic resonance imaging.