This study examined the reliability and validity of a behavioral observation method for the assessment of arthritis pain in a clinical practice setting. Trained observers measured the occurrence of seven pain behaviors in a group of 61 rheumatoid arthritis patients undergoing physical examinations. These observations were compared with videotaped observations of the patients in a laboratory setting. Significant differences were found between the pain behavior frequencies observed during the examinations and those observed during videotaped sessions. Total pain behavior scores obtained in both settings were significantly correlated with patients' self-reports of pain and with disease activity measures. Pain behavior observed during the exams was significantly associated with patients' self-reports of anxiety and depression.