Social constructivist theories of race suggest no two measures of race will capture the same information, but the degree of "error" this creates for quantitative research on inequality is unclear. Using unique data from the General Social Survey, I find observed and self-reported measures of race yield substantively different results when used to explain income inequality in the United States. This occurs because inconsistent racial classification is correlated with other respondent characteristics such as immigrant generation, educational attainment and age. (Contains 4 tables, 2 figures and 9 notes.)