To assess the utility of relatives' assessments of pain in cognitively impaired nursing home residents, the internal consistency of these assessments, and their relationship to other assessments of pain, was examined in a correlational study of 79 residents of a large suburban nursing home. The cohort was 85% female and had an average age of 87 years. The sample included moderately and severely cognitively impaired residents, some of whom were taking pain medication. The results demonstrated that relatives were less likely to rate the pain of the resident when the resident's cognitive level was more impaired and when they had a longer stay in the nursing home. The internal consistency of the relatives' ratings were good and, when only relatives who visited at least once a week were included, correlated significantly with most Minimum Data Set (MDS), resident, physician, and nursing staff ratings. They also correlated significantly with frequency of visits, higher cognitive function, type of relationship with the resident, and with a shorter stay in the nursing home. Relatives' ratings of pain and of past sources of pain may prove useful in the detection of pain in cognitively impaired persons. However, relative's pain ratings are only useful when relatives visit regularly. Like other informants, relatives have more difficulty rating pain when the resident is severely cognitively impaired.