In this paper we abstract from the usual gains and costs of decentralization (e.g. preference matching, spillovers and economies of scale). Instead we compare the political accountability of decentralized governments relative to centralized ones when there is a risk of "bad" governance. We study both the selection and incentive effects of accountability. A key aspect of centralization is to make the politician answerable to multiple constituencies subject to a common budget constraint. Our main findings are that (a) when politicians differ in competence, decentralization unambiguously dominates; and (b) when politicians differ in honesty, decentralization and centralization have conflicting accountability effects (when one provides better discipline, the other gives better selection). The analysis then identifies the circumstances under which centralization may increase voter welfare. The more general lesson that we can draw is that different institutional forms give rise to different information to the voters on which electoral accountability can be based. Therefore they differ on how effective elections can be in disciplining and selecting policymakers.