In the first part of this article, we address the problems inherent in studying wrongful convictions: our pervasive ignorance and the extreme difficulty of obtaining the data that we need to answer even basic questions. The main reason that we know so little about false convictions is that, by definition, they are hidden from view. As a result, it is nearly impossible to gather reliable data on the characteristics or even the frequency of false convictions. In addition, we have very limited data on criminal investigations and prosecutions in general, so even if we could somehow obtain data on cases of wrongful conviction, we would have inadequate data on true convictions with which to compare them. In the second part of the article, we dispel some of that ignorance by considering data on false convictions in a small but important subset of criminal cases about which we have unusually detailed information: death sentences. From 1973 on, we know basic facts about all defendants who were sentenced to death in the United States, and we know which of them were exonerated. From these data we estimate that the frequency of wrongful death sentences in the United States is at least 2.3 percent. In addition, we compare post-1973 capital exonerations in the United States to a random sample of cases of defendants who were sentenced in the same time period and ultimately executed. Based on these comparisons, we present a handful of findings on features of the investigations of capital cases, and on background facts about capital defendants, that are modest predictors of false convictions.