Prospective evidence for predictors of decline in social relationships over a 3-year follow-up period in an elderly cohort are presented. The cohort consisted of men (n = 903) and women (n = 1,673) over age 65 years in two rural Iowa counties who were interviewed in 1982 and again in 1985. Three separate measures of social relationships were dichotomized into lower and higher levels and included the number of close friends and relatives (less than three vs. three or more), church attendance (less than once per month or not at all vs. once per month or more), and membership in a group (nonmember vs. member). Those with higher social relationship levels at both interviews were compared with those who had higher levels at baseline but lower levels at follow-up (i.e., a decline in social relationship level) using logistic regression. In multivariate analysis, important baseline predictors of decline in social relationship levels included greater age, lower educational level, lower memory test score, the presence of physical disabilities, and a higher level of depressive symptoms. Marital status and lower self-perceived health status were less consistent predictors, and having any living children, history of major illness, and continence status were generally not important predictors of decline in social relationship levels. These findings underscore the multifactorial and complex influences on changes in social relationships, but they also identify factors for possible prevention and intervention strategies.