Four studies examined the reliability and validity of a behavioral observation method for the assessment of pain associated with rheumatoid arthritis (RA). The major purpose of experiment 1 was to evaluate the interobserver reliability of the observation method. Two observers recorded the frequencies of pain behaviors displayed by 20 RA patients. Each of 3 types of reliability estimates indicated that the pain behavior could be reliably observed. The purpose of experiment 2 was to examine the concurrent validity of the observation system by correlating 53 patients' self-reports of pain with the frequencies of their pain behaviors. Significant and positive correlations were found between patients' total pain behavior and 3 self-report measures of pain and functional disability. Furthermore, unlike the self-reports of pain, total pain behavior was only minimally related to self-report of depression. Experiment 3 was performed in order to assess the observation method's construct validity. Naive observers viewed video recordings of 25 RA patients and made global estimates of patients' pain severity and unpleasantness. Highly significant and positive correlations were found between these global estimates and total pain behavior. In experiment 4, the pain behaviors of 11 RA patients were recorded prior to and immediately following cognitive-behavioral treatment for the reduction of RA pain. There was a significant decrease in total pain behavior from pretreatment to posttreatment. The data indicate that the observation method provides a reliable, valid, and relatively objective measure of RA patient pain behavior. Future validation studies of the observation method are discussed.