Previous research has demonstrated that equivalence relations are difficult to reverse. The present study sought to determine if this difficulty is related to particular features of the training tasks. Sixty-six Dutch adults and children were employed across 7 experiments. Following the formation of two 3-term equivalence classes (A1-B1-C1, A2-B2-C2), children in Experiment 1 received reversal training followed by equivalence probes. Two reversal training procedures were compared: standard and nonstandard. In standard A-B reversal training, the subjects had to choose between the same 2 different-class comparisons as in baseline (B1 vs. B2, C1 vs. C2). This training requires different choices during the A-B tasks but not during the A-C tasks. In nonstandard A-B reversal training, the subjects had to choose between 2 same-class B and C stimuli (B1 vs. C1, B2 vs. C2). This procedure requires the subjects to make novel choices during all training tasks. Almost all subjects showed reversal of equivalence relations irrespective of the training protocol. Experiments 2-7 examined if this finding could be related to the subject's age and mental ability, the response topography, the identity and location of the experimenter, specific features of the training and testing program that were different from those in previous research (e.g., Pilgrim & Galizio, 1990), or to the subject's ability to accurately predict the probability of reinforcement. Unless the probes were introduced before the reversal training had been completed and the nonstandard protocol was used, both reversal protocols almost always produced equivalence reversal.