One of the most serious criticisms of research on the contact hypothesis is the contention that selection bias operates to promote interaction between whites and blacks who are already relatively unprejudiced toward one another. Accordingly, attempts to infer the effect of interracial contact on racial attitudes must recognize and correct for this potential source of bias. Endogenous switching regression models are used to estimate the effect of close interracial friendship on selected racial attitude variables while accounting for possible selectivity bias. Each model implies a distinct assumption about the process generating interracial contact and racial attitudes. Using data from the National Survey of Black Americans, we find no evidence of sample selection bias in estimating the effect of close interracial contact on black racial attitudes. Therefore, we cannot reject the conventional models used in research on the contact hypothesis that treat interracial contact as an exogenous variable affecting racial attitudes. However, our results show the usefulness of a switching regressions approach to reveal the contingent nature of the contact hypothesis.