Rats responded in a six‐stimulus, two‐response temporal classification procedure. A successive‐reversal design was used in which the relationship between stimulus class (short vs. long) and correct comparison location (left or right) reversed every 15 sessions. After several reversals, the relative probability of reinforcement for each correct classification was manipulated across subsequent reversals. In each condition, the asymptotic level of preference for the comparison location (response bias) correlated with the greater probability of reinforcement was demonstrated in the first session following a reversal, whereas discrimination accuracy took several more sessions to return to asymptotic levels. A modified version of the attending‐augmented Davison‐Nevin‐Alsop (Davison & Nevin, 1999) model offered by Nevin, Davison, & Shahan (2005) provided an accurate description of the reacquisition data. The comparison‐attending parameters remained high and relatively constant following reversals, while sample‐attending parameters initially decreased following reversals, and then increased gradually across sessions. These findings support key assumptions of the attending model; sample‐ and comparison‐attending are independent processes that modulate the expression of discriminative control exerted by those stimuli over operant behavior.