As part of the Alzheimer's Disease Cooperative Study (ADCS) Instrument Development Project, the CERAD Behavior Rating Scale for Dementia (BRSD) was examined for its sensitivity to degree of cognitive impairment, its test-retest reliability, and its sensitivity to longitudinal change. Sixty-four normal elderly participants and 261 patients with AD stratified into severity groups based on Mini-Mental State Exam (MMSE) scores were rated on the BRSD at baseline and 12-month follow-up visits. A subset of subjects was evaluated at a 1-month follow-up visit. Baseline BRSD Total Score discriminated the normal group from each AD group, and mean Total Score significantly increased with increasing dementia severity. Test-retest reliability between baseline and 1-month Total Scores was satisfactory for all AD groups. Longitudinal change was evaluated by 12-month change scores, which were significant in only the normal and in one AD group. From the results, we argue that the value of behavioral pathology assessment in clinical trials would be enhanced if additive scores were based on groups of correlated items rather than on a broad array of behaviors, some of which may increase and others may decrease in frequency as AD progresses.