Participants were trained in a series of interrelated conditional discriminations that aimed to establish four 4-member equivalence classes (i.e., A1–B1–C1–D1, A2–B2–C2–D2, A3–B3–C3–D3, A4–B4–C4–D4). During this training, the four A stimuli (i.e., A1, A2, A3, and A4) were compounded with pictures containing positive or negative evaluative functions (A1/A2 negative & A3/A4 positive). The transfer of evaluative functions to directly and indirectly related members of the equivalence classes (i.e., B, C, and D stimuli) was measured using an Implicit Association Test (IAT). During consistent test blocks, participants were required to press the same response key for target words that were related to those A stimuli that possessed similar evaluative functions (A1/A2-left key & A3/A4-right key). During inconsistent test blocks, target words that were related to those A stimuli with different evaluative functions were assigned to the same response key (A1/A4–left key & A2/A3–right key). Results showed that all 8 participants, who passed a matching-to-sample equivalence test following the IAT, responded more rapidly on consistent relative to inconsistent test blocks. This typical IAT effect was not observed for those participants who did not pass the equivalence test. The results suggest that the IAT effect may arise from formally untested derived relations, and supports the argument that such relations could provide a valid behavioral model of semantic categories in natural language.