The purpose of this study is twofold: 1) to model changes in health over time among older adults; and 2) to assess the degree to which stress arising in salient social roles accounts for individual variation in these changes. Individual growth curve analyses using Hierarchical Linear Modeling (HLM) software were employed with longitudinal data collected from a nationwide sample of 465 older adults between 1992-1999. The data suggest that, throughout the course of the study, the sample as a whole experienced linear deterioration in self-rated health. After accounting for the effects of non-salient role stress and demographic characteristics, variance in health changes was not accounted for significantly by salient role stress in an additive model. However, the inclusion of a multiplicative term showed that as respondent age increased, the association between salient role stress and adverse changes in health strengthened significantly. These results support the idea that the oldest-old may be particularly vulnerable to the effects of stress. This finding has important implications for targeting interventions and should stimulate further study regarding how and why responses to stress change progressively with age. (Contains 2 tables.)