The purpose of this study was to assess the reliability and validity of facial expressions as pain indicators in patients with severe dementia. Based on interviews with patients who could report pain, we defined characteristics of decubitus ulcers associated with reports of pain during dressing changes. We then evaluated 9 patients who had ulcers with these characteristics but were unable to communicate verbally because of severe dementia. We videotaped their facial expressions before and during their decubitus ulcer dressing change. We showed the videotape segments, in random order, to 8 medical students and 10 nurses. The 18 viewers were asked to infer the presence or absence of pain based on their observations of the patients' facial expressions and vocalizations. The dressing change of decubitus ulcers extending beyond the subcutaneous tissue, covering an area of at least 9 cm(2), and with a moist surface, was always reported as painful by study patients able to report (95% confidence interval of 69-100%). The intraclass correlation coefficient for the answers of the 18 viewers evaluating each videotape segment for the presence of pain was 0.64. Sensitivity, specificity, and positive and negative predictive values of viewers' ratings of facial expressions and vocalizations as a measure of the presence of pain were: 0.70, 0.83, 0.90, and 0.81. The intraclass correlation coefficient for the answers rating pain intensity was only 0.10, indicating only slight agreement beyond chance. Assuming dressing changes of ulcers reported as painful by communicative patients are also painful in non-verbal severely demented patients, clinician observations of facial expressions and vocalizations are accurate means for assessing the presence of pain, but not its intensity, in patients unable to communicate verbally because of advanced dementia.