Studies of the behavioral symptoms in Alzheimer's disease (AD) would be facilitated if reliable reports could be obtained retrospectively, especially about symptoms in the final months of life when concurrent assessment is often not feasible. To evaluate such a method, we compared results of a telephone interview conducted after the patient's death with information provided by the same informant earlier, while the patient was living. Agreement between in-life and retrospective assessments was higher for psychotic symptoms than for depressive behaviors, suggesting that retrospective assessment of specific behavioral symptoms in AD is not uniformly reliable. More symptoms were retrospectively reported as the length of time between the last in-life evaluation and date of death increased, suggesting that there may be increased variety of behavioral symptoms observed by caregivers in the final stages of the disease.