The model of tax structure developed in this paper is one in which the composition of revenues and the structure of specific taxes arise 'naturally' as the result of self-interested political behavior. It is assumed that political agents choose tax structure so as to minimize the political costs (or expected net loss in votes) associated with raising a budget of given size. Several hypotheses concerning the nature of political cost functions are developed and applied to an explanation of the differences among U.S. states in their reliance on income taxation. The empirical application emphasizes differences in political constraints across jurisdictions in the belief that much can be learned about the choice of policy instruments by studying structural adjustments in response to varying constraints. The paper discusses how the proposed approach differs from or is consistent with that adopted by other authors, and how the estimating equation relates to other empirical research in the literature.