To investigate the clinical variables that contribute to the physician's opinion of the medium term outcome of rheumatoid arthritis (RA), we compared the physician's opinion with statistical modelling of these clinical variables. In 133 young female patients with RA followed yearly (mean followup duration 6 years) from symptom onset, the physician had to state whether the disease course was mild or progressive. Using logistic regression models we concluded that the physician's opinion could be explained by using a compilation of the erosion score, the number of swollen joints, the Ritchie score as measured at the latest visit, in combination with either the mean number of swollen joints during followup, or the number of prescribed second line drugs. Several different models with a common core were equally close to the physician's opinion.