Harsanyi (1955) showed that the only way to aggregate individual preferences into a social preference which satisfies certain desirable properties is ``utilitarianism'', whereby the social utility function is a weighted average of individual utilities. This representation forms the basis for welfare analysis in most applied work. We argue, however, that welfare analysis based on Harsanyi's version of utilitarianism may overlook important distributional considerations. We therefore introduce a notion of utilitarianism for discrete-choice settings which applies to \textit{social choice functions}, which describe the actions of society, rather than social welfare functions which describe society's preferences (as in Harsanyi). We characterize a representation of utilitarian social choice, and show that it provides a foundation for a family of \textit{distributional welfare measures} based on quantiles of the distribution of individual welfare effects, rather than averages.