The research examines an unobtrusive measure of racial attitudes based on the evaluations that are automatically activated from memory on the presentation of Black versus White faces. Study 1, which concerned the technique's validity, obtained different attitude estimates for Black and White participants and also revealed that the variability among White participants was predictive of other race-related judgments and behavior. Study 2 concerned the lack of correspondence between the unobtrusive estimates and Modern Racism Scale (MRS) scores. The reactivity of the MRS was demonstrated in Study 3. Study 4 observed an interaction between the unobtrusive estimates and an individual difference in motivation to control prejudiced reactions when predicting MRS scores. The theoretical implications of the findings for consideration of automatic and controlled components of racial prejudice are discussed, as is the status of the MRS.