OBJECTIVE To evaluate regression models that include social, attitudinal, work structure, health status, and family characteristics, with regard to their prediction of work disability in a national sample of patients with rheumatoid arthritis (RA). METHODS Four hundred ninety-eight employed RA patients were recruited from a national sample of private rheumatology practices. Three hundred ninety-two remained in the study after 5 years. Data were collected from patients by telephone interview, and patients' physicians provided written clinical assessments. Only variables on which information was obtained in year 1 were used to predict work status in year 5, using hierarchical multiple logistic regression analysis. RESULTS The significant predictors of work disability were age (odds ratio [OR] 1.04), number of deformed joints (OR 1.26), number of joints with flare (OR 1.23), the complexity of working with things at work (OR 0.88), and the desire to remain employed (OR 2.3). The risk of work disability increased with increasing age, more severe disease, greater complexity of involvement with things at work, reduced work hours, and desire to not be working outside the home. CONCLUSIONS The risk of becoming work disabled in 5 years was predicted more by clinical status at entry into the study than by work structure. These results, which contradict previous research on work disability in arthritis, prompt a rethinking of future studies of work disability in RA.