Speech patterns differ substantially between whites and many African Americans. I collect and analyze speech data to understand the role that speech may play in explaining racial wage differences. Among blacks, speech patterns are highly correlated with measures of skill such as schooling and AFQT scores. They are also highly correlated with the wages of young workers. Even after controlling for measures of skill and family background, black speakers whose voices were distinctly identified as black by anonymous listeners earn about 12 percent less than whites with similar observable skills. Indistinctly identified blacks earn essentially the same as comparable whites. I discuss a number of models that may be consistent with these results and describe the data that one would need to distinguish among them. (Contains 10 tables and 11 footnotes.)