Scholars have long focused on the risks of poverty, defined as individual-level labor market and family characteristics more common among the poor than the non-poor. This talk first develops a framework for analyzing the risks of poverty in terms of prevalences (share of the population with a risk) and penalties (increased probability of poverty associated with a risk). Using the Luxembourg Income Study, Brady will compare the prevalences and penalties of the four chief risks (low education, single motherhood, young headship, and unemployment) across 29 rich democracies. There is much greater cross-national variation in penalties than prevalences. Second, Brady will apply this framework to the U.S. He will show that the unusually high U.S. poverty results from very high penalties despite below average prevalences. Counterfactual simulations demonstrate U.S. poverty would decline more with cross-national median penalties than cross-national median prevalences. Moreover, U.S. poverty in 2013 would actually be worse with prevalences from the U.S. in 1970 or 1980. Third, he will analyze the cross-national variation in prevalences and penalties. He finds little evidence of the expected negative relationship between penalties and prevalences. He will also show welfare generosity significantly moderates the penalty for two risks: unemployment and low education. Among other conclusions, we propose: a) existing variation in risks cannot explain most of the variation in poverty; and b) studies of the risks of poverty based solely on the U.S. are constrained by sample selection biases.