Previous studies of racial inequality have relied on official statistics that presumably use self-classification of race. Using novel data from a 1995 national survey in Brazil, we find that the estimates of racial income inequality based on self-classification are lower than those based on interviewer classification. After human capital and labor market controls, whites earn 26% more than browns with interviewer classification but earn only 17% more than browns with self-classification. Black-brown differences hardly change: Blacks earn 13% and 12% less than browns with interviewer classification and self-classification, respectively. We contend that interviewer classification of race is more appropriate because analysts of racial inequality are interested in the effects of racial discrimination, which depends on how others classify one's race.