Using a Latin square design, 4 patients were examined by 4 rheumatologists. Joints were scored for tenderness and inflammation. The Ritchie, the index of the American Rheumatism Association (ARA), the Hart modified Ritchie and a simplified Lansbury index were calculated from the raw data. The results suggest that an articular index consisting of a simple count of tender joints (Hart modified Ritchie) or a simple count of tender or swollen joints (ARA index) are the most reproducible with multiple observers. We suggest that these indices would be most appropriate for multicenter clinical trials.