Agroforests managed by smallholders have been shown to provide biodiversity, carbon-storage and rural-livelihood services. Consequently, these systems are being promoted as an effective way of rehabilitating millions of hectares of degraded, formerly forested land in many tropical countries. Current conditions at the forest margins in these countries, however, make it easier to clear unprotected forests than restore degraded lands through agroforestry. The result is large-scale deforestation that causes substantial losses of biodiversity and stored soil and biomass carbon. Agroforests will only be an attractive activity if they are financially viable and socially acceptable. In this study we investigate the financial viability of agroforestry systems as carbon sinks when carbon-credit payments are available. A meta-modelling framework is adopted, comprising an econometric-production model of a land parcel in Sumatra, Indonesia. The model is used within a dynamic-programming algorithm to determine optimal management of the system in terms of three decision variables: tree/crop area, tree-rotation length, and wood harvest. Results show the influence of soil-carbon stocks and discount rates on optimal strategies and reveal interesting implications for joint management of agriculture and carbon as well as for the possible restoration of degraded land.