This research examines longitudinal data on psychosocial status and polysomnographic sleep collected annually from 57 healthy, community-residing elders aged 61-89. Cluster analysis of variables reflecting sleep continuity and architecture at the baseline assessment was used to identify three groups of elders: those whose sleep was either (a) superior to all remaining respondents across a variety of measures, (b) marred only by significantly reduced sleep efficiency relative to other respondents, or (c) poorer than all other respondents in multiple areas. Cross-validation procedures suggested that the three-group cluster solution was stable and replicable over persons and over time. Subsequent multivariate analyses indicated that recent life events, as well as psychosocial stability and support variables at baseline, distinguished between the sleep pattern groups. Moreover, sleep pattern group membership itself predicted subjects' subsequent sleep characteristics and psychosocial status at follow-up. Implications of these results for conceptualizing psychosocial factors that affect, and are ultimately affected by, sleep disturbances in late life are discussed.