After blogging some simple regressions showing that, contra Harford, blacks actually have an unusually high return to education, I emailed my friend, and noted labor economist, Gordon Dahl, for further background. He gave me permission to reprint the following:
Neal and Johnson (1996, JPE) was the seminal paper I was referring to in my last email. It actually focuses on the overall black-white wage gap (not the gap in the return to education between blacks and whites). The main point there is that much of the black-white wage gap for men, and all of the gap for women, disappears if you add afqt as an additional covariate in a log earnings regression. Interestingly, they find no significant difference in the return to afqt by race. If you look at Appendix Table 1, you do find that blacks earn a significantly higher return for a college degree compared to whites. This is the type of result I presume you were referring to in your previous email.
Oettinger (1996) has a very nice paper on statistical discrimination which also interacts black with education. While he finds a positive estimated coefficient on the interaction, it isn't significant in his specification.
More generally, when researchers do Oaxaca-type decompositions, they often break things into the portion attributable to differences in characteristics between blacks and whites versus differences in the returns to those characteristics (see Altonji and Blank's 1999 handbook chapter for examples and an excellent summary of what was known about race and the labor market more generally). The estimated coefficients are not always reported in such papers, but researchers definitely know when differences exist, since they need to use this information to calculate the differences.
Bottom line: The NLSY regressions I posted seem unusually strong and clear-cut, but the basic result that blacks have a larger-than-average return to education is already known in the literature.