An appetitive conditioning experiment with rats assessed the predictions of a new performance-based account of associative learning called the computational comparator hypothesis (Murphy, Baker, & Fouquet, 2001a, 2001b). A between-subjects design was used in which the stimuli A or B were separately trained either as excitors or as inhibitors prior to and during a relative validity treatment. During relative validity training, X was reinforced when presented with A but was not reinforced when presented with B. In test, responding to X in extinction was lower when A or B had been separately trained as excitors than as inhibitors. Thus, contrary to the computational comparator hypothesis, responding to X was affected by more than just inhibitory training of A. Better fits to the data were obtained by Pearce’s configural theory (Pearce, 1987, 1994) and the extended comparator hypothesis (Denniston, Savastano, & Miller, 2001) than by the elemental theory of Rescorla and Wagner (1972) or the computational comparator hypothesis.