In a context where clientelism is widespread, why do some politicians choose not to condition the delivery of goods and services to citizens on individual political behavior? I argue that the answer to this question lies in the heretofore unexamined electoral costs of clientelism: clientelism decreases support from nonpoor constituents even while it generates votes from among the poor. Taking into account these costs and other factors that shape politician incentives, I posit that the interaction between political competition and poverty will explain variation in clientelism. I test this claim using an original measure of clientelism that assesses mayoral involvement in social policy implementation in Argentine municipalities. The results of statistical analysis suggest that high levels of political competition are compatible with clientelism when poverty is also high. Only when high competition is coupled with low rates of poverty does clientelism decline.