The purpose of the study was to compare the psychometric properties of four established pain scales in a population of hospitalized older adults (mean age, 76 years) with varying levels of cognitive impairment. Patients made ratings of current pain three times/day for 7 days. They also made retrospective daily, weekly, and bi-weekly ratings of usual, worst, and least pain levels over a 14-day period. Ratings were made on four different scales, varying in numeric and verbal demands: a five-point verbal rating scale, a seven-point faces pain scale, a horizontal 21-point (0-100) box scale, and two vertical 21-point (0-20) box scales (measuring pain intensity and pain unpleasantness). The accuracy, reliability, construct validity, postdictive validity, and bias susceptibility of each scale were evaluated. The horizontal 21-point box scale emerged as the best scale with respect to both psychometrics and validity, regardless of mental status. Pain intensity did not vary as a function of mental status. Retrospective estimates of pain varied by mental status: a combination of usual/worst pain was best for cognitively impaired patients, while a combination of usual/least pain was best for unimpaired patients. These findings support the use of the 21-point box scale for pain assessment in older patients, including those with mild-to-moderate cognitive impairment. They also support the ability of older, cognitively impaired patients to rate pain reliably and validly.