Our objective was to compare the validity and reliability of three formats for self-administered articular indices assessing pain (PAI) or swelling (SAI). Fifty-five patients with rheumatoid arthritis were asked to mark the degree of pain on a list of 16 joints (PAI list), to mark 'painful joints' on a mannequin presenting 42 joints (PAI diagram), and to mark 'swollen or tender joints' on a mannequin presenting 38 joints (SAI diagram). The test--retest reliability (intraclass correlation coefficient) ranged from 0.63 (SAI diagram) to 0.67 (PAI diagram) and 0.85 (PAI list). The correlation with clinical parameters was strongest for the PAI list and the SAI diagram. The association of the SAI diagram with clinical parameters increased with omission of the less reliable toe joints and/or weighting for joint size according to Lansbury. As expected, the short and weighted SAI diagram correlated more strongly with the physician-derived swollen joint count (r = 0.49), C-reactive protein (r = 0.49) and erythrocyte sedimentation rate (r = 0.41) than did the PAI list whereas the PAI list correlated more strongly with physician-derived tender joint count (r = 0.43), global pain measured on a numerical rating scale (r = 0.57) and the Health Assessment Questionnaire (r = 0.49) than did the SAI diagram. We concluded that patients' rating of tender and swollen joints on a mannequin diagram and calculation of a 26-joint and weighted articular index produces an excellent estimate of total joint inflammation, which may be useful in clinical, health services and epidemiological research. An articular index calculated from ratings of pain degree of 16 joints or joint groups may provide complementary information.