Three experiments investigated the symbolic generalization (transformation) of discriminative (bar-pressing) functions in accordance with a five-member arbitrary relational comparative network (A–B–C–D–E) in adults. Following nonarbitrary relational training and testing to establish the contextual functions of MORE-THAN and LESS-THAN for two abstract images, participants received arbitrary relational training and testing. One group received training on “More-than” baseline relations (e.g., B > A, C > B, D > C, and E > D) and another group received training on “Less-than” baseline relations (A A and A < C, etc.) and exposed to a bar-press training phase, which trained a low, steady rate bar-pressing response function to the middle stimulus (C) in the relational network. In Experiment 1, testing involved a quasi-random order of presentation of probe stimuli and half of the participants responded in accordance with the predicted relational network (i.e., pressed less to A and B and more to D and E, than to C). In Experiment 2, none of the participants showed the predicted performance when the probe stimuli were presented in a fixed order (A–B–C–D–E), whereas three out of four participants in Experiment 3 responded in accordance with the predicted relational network when the test was repeated. These findings indicate the importance of identifying methodological factors that may potentially influence the symbolic generalization of discriminative functions.