We collect a new dataset on capital punishment in the United States and we propose a test of racial bias based upon patterns of sentence reversals. We model the courts as minimizing type I and II errors. If trial courts were unbiased, conditional on defendant's race the error rate should be independent of the victim's race. Instead we uncover 3 and 9 percentage points higher reversal rates in direct appeal and habeas corpus cases, respectively, against minority defendants who killed whites. The pattern for white defendants is opposite but not statistically significant. This bias is confined to Southern states. (JEL J15, K41, K42) One of the arguments against the death penalty in the United States is that it is applied with a racial bias against minorities. Consider for example the following statement, taken from the opening paragraph of a document by one of the most vocal organizations opposing capital punishment:African Americans are disproportionately represented among people condemned to death in the USA. While they make up 12 percent of the national population, they account for more than 40 percent of the country's current death row inmates, and one in three of those executed since 1977.