This paper reports the frequency and predictors of remission (no arthritis on examination and no treatment with second-line drugs or steroids within the previous 3 months) in 358 patients with early inflammatory polyarthritis (IP) referred to the Norfolk Arthritis Register. Two years after referral, 91 patients (25%) were in remission, 32 of whom had also been in remission at 1 yr. Remission rates were twice as high in patients with undifferentiated inflammatory polyarthritis at baseline as in those who satisfied criteria for rheumatoid arthritis. To identify predictors of remission, a logistic regression model was developed on a random two-thirds of the patients and validated on the remaining one-third. Remission at 2 yr was associated with male gender and fewer than six tender joints at baseline. However, even the best-fitting model was not sensitive enough to be useful clinically. Thus, amongst patients with early IP in the community, remission rates at 2 yr are low. Further, it was impossible, using simple clinical measures, to predict those patients whose arthritis would resolve.