Deposit insurers are particularly concerned about high-cost failures. When the factors driving such failures differ systematically from the determinants of low- and moderate-cost failures, a new estimation technique is required. Using a sample of more than 1,000 bank failures in the U.S. between 1984 and 2003, I present a quantile regression approach that illustrates the sensitivity of the dollar value of losses in different quantiles to my explanatory variables. These findings suggest that reliance on standard econometric techniques results in misleading inferences, and that losses are not homogeneously driven by the same factors across the quantiles. I also find that liability composition affects time to failure.