OBJECTIVES: To assess the performance of self-assessment scales in severely demented hospitalized patients and to compare it with observational data. DESIGN: Prospective clinical study. SETTING: Geriatrics hospital and a geriatric psychiatry service. PARTICIPANTS: All patients who met Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, criteria for dementia, with a Mini-Mental State Examination score less than 11 and a Clinical Dementia Rating score of 3. MEASUREMENTS: Three self-assessment tools--the verbal, horizontal visual, and faces pain scales--were administered in randomized order. A nursing team independently completed an observational pain rating scale. Main outcomes were comprehension (ability to explain scale use and correctly indicate positions for no pain and extreme pain, on two separate occasions), inter- and intrarater reliability, and comparison of pain intensities measured by the different scales. RESULTS: Sixty-one percent of 129 severely demented patients (mean age 83.7, 69% women) demonstrated comprehension of at least one scale. Comprehension rates were significantly better for the verbal and the faces pain scales. For patients who demonstrated good comprehension, the inter- and intrarater reliability of the three self-assessment scales was high (intraclass correlation coefficient=0.88-0.98). Correlation between the three self-assessment scales was moderate to strong (Spearman correlation coefficient (r)=0.45-0.94; P<.001). Observational rating correlated at least moderately with self-assessment (r=0.25-0.63), although for patients reporting pain, the observational rating scale underestimated severity compared with all three self-assessment scales. CONCLUSION: Clinicians should not apply observational scales routinely in severely demented patients, because many are capable of reliably reporting their own pain.