This study aimed to verify the role of three parameters on the formation of equivalence classes between Black faces and a positive symbol, in children who demonstrated negative bias toward Black faces in a pretest. Maintenance was also verified 6 weeks after equivalence tests. Forty-six children (11 Black; 27 girls) who demonstrated racial bias in a pretest were divided into four groups. All groups first learned AB relations (A1 and A2 were, respectively, a positive and a negative symbol, and B were abstract stimuli) and then BC relations (C1 was a Black face and C2 was an abstract stimulus). The Control Group then advanced immediately to equivalence tests (AC, and CA, without differential consequences). For the Mixed Training Group, a block of trials mixing AB and BC relations, with differential consequences, preceded equivalence tests. For the Feedback Reduction Group, equivalence tests were preceded by a trial block mixing AB and BC relations, but with feedback in 50% of trials. The Symmetry Group received symmetry tests after training of each baseline relation. Thirty-three children showed class formation relating Black faces and the positive symbol, and 27 maintained at least one of the equivalence relations after 6 weeks. Average biases toward Black faces were positive in a posttest, for participants who formed equivalence classes, and remained negative for those that did not form classes. The Control Group showed less pronounced bias reduction and maintenance of relations after 6 weeks, suggesting that these outcomes may be affected by training parameters.