Decentralization has long been advocated as a desirable process for improving health systems. Nevertheless, we still lack a sufficient analytical framework for systematically studying how decentralization can achieve this objective. We do not have adequate means of analyzing the three key elements of decentralization: (1) the amount of choice that is transferred from central institutions to institutions at the periphery of health systems, (2) what choices local officials make with their increased discretion and (3) what effect these choices have on the performance of the health system. This article proposes a framework of analysis that can be used to design and evaluate the decentralization of health systems. It starts from the assumption that decentralization is not an end in itself but rather should be designed and evaluated for its ability to achieve broader objectives of health reform: equity, efficiency, quality and financial soundness. Using a "principal agent" approach as the basic framework, but incorporating insights from public administration, local public choice and social capital approaches, the article presents a decision space approach which defines decentralization in terms of the set of functions and degrees of choice that formally are transferred to local officials. The approach also evaluates the incentives that central government can offer to local decision-makers to encourage them to achieve health objectives. It evaluates the local government characteristics that also influence decision-making and implementation at the local level. Then it determines whether local officials innovate by making choices that are different from those directed by central authorities. Finally, it evaluates whether the local choices have improved the performance of the local health system in achieving the broader health objectives. Examples from Colombia are used to illustrate the approach. The framework will be used to analyze the experience of decentralization in a series of empirical studies in Latin America. The results of these studies should suggest policy recommendations for adjusting decision space and incentives so that localities make decisions that achieve the objectives of health reform.