We study distributional preferences in “large” groups. While most prior experiments have focused on exploring attitudes toward inequality in two- or three-person groups, we field a series of experiments via Mechanical Turk in which subjects choose between two income distributions, each with seven (or nine) individuals, with hypothetical incomes that aim to approximate the actual distribution of income in the U.S. Our setting thus provides a more direct comparison to the redistributive choices faced by society. Consistent with standard maximin (Rawlsian) preferences, subjects select distributions in which the bottom individual’s income is higher (but show little regard for lower incomes above the bottom ranking). In contrast to standard models, however, we find that subjects select distributions that lower the top individual’s income, but not other high incomes. Finally, we provide tentative evidence of “locally competitive” preferences–in most experimental sessions, subjects select distributions that lower the income of the individual directly above them, while the income of the individual two positions above has little effect on subjects’ decisions. Our findings suggest that theories of inequality aversion should be enriched to account for individuals’ aversion to “topmost” and “local” disadvantageous inequality.