This two-part paper provides a theoretical framework for appraising trade-offs between alternative methods of delegating authority over the delivery of public services, on the targeting and cost-effectiveness of public spending programs in developing countries. Authority over these programs has to be delegated owing to absence of information at the central level concerning local needs and costs of specific communities. In a top-down centralized system, this authority is delegated to bureaucrats by a central government that has limited ability to monitor their performance with respect to either service delivery or cost control. In a decentralized system, it is allocated instead to elected local governments or client groups, which may be subject to capture by local elites. Both systems are thus prone to local corruption and lack of accountability. Part 1 of the paper studies the relevant tradeoffs in the context of a poverty alleviation program, whose aim is to deliver a private merit good available on competitive markets to the poor. Decentralization generally dominates with respect to inter-community targeting as well as cost-effectiveness. However, the ranking of intracommunity targeting under the two systems is ambiguous, and depends on the relative degree of capture that local and national governments are prone to, besides the nature of uncertainty and preferences of the good by the nonpoor. Part 2 of the paper considers an infrastructure service provided by a public enterprise which has a natural monopoly. In this context it is shown that decentralization dominates if the following four conditions are satisfied: (i) local governments are not vulnerable to capture; (ii) local governments have access to adequate local financing sources; (iii) there are no interjurisdictional externalities in service provision; and (iv) local governments have all the bargaining power and access to relevant cost information vis-a-vis public enterprise managers. Absent any one of these institutional conditions, however, decentralization may perform worse than centralization. The Appendix develops a model of electoral competition (adapted from Grossman-Helpman (1996)) where parties are prone to capture by special interest groups, which helps identify some of the institutional determinants of the degree of capture of local and central governments.