The impacts of the productive social activities of volunteer and paid work on health have rarely been investigated among the oldest Americans despite a recent claim for their beneficial effect (Rowe and Kahn 1998). This paper used data from Waves 3 and 4 of the Asset and Health Dynamics among the Oldest Old (AHEAD) Study to (1) investigate the impact of these activities on health measured as self-reported health and activities of daily living (ADL) functioning limitations and to (2) explore possible causal mechanisms. Using multinomial logistic regression analysis, amounts of volunteer and paid work over a minimum of 100 annual hours self-reported at Wave 3 were related to poor health and death as competing risks measured at Wave 4, controlling for health measured at Wave 2 and for other predictors of poor health and death. Findings suggest that performing more than 100 annual hours of volunteer work and of paid work have independent and significant protective effects against subsequent poor health and death. Additional analyses suggest that the quantity of volunteer and paid work beyond 100 annual hours is not related to health outcomes and that physical exercise and mental health measured as cognitive functioning and depressive symptoms explain not entirely overlapping parts of the relationship between productive activities and health.