Mirror generalization is detrimental for identifying letters with lateral mirror-image counterparts ('b/d'). In the present study, we investigated whether the discrimination of this type of letters in expert readers might be rooted in the ability to inhibit the mirror-generalization process. In our negative priming paradigm, participants judged whether two letters were identical on the prime and two animals (or buildings) were identical on the probe. In Experiment 1, participants required more time when determining that two animals (but not two buildings) were mirror images of each other when preceded by letters with mirror-image counterparts than without mirror-image counterparts ('a/h'). In Experiment 2, we replicated the results with different letters without mirror-image counterparts and with the type of probe stimuli (animal or building) manipulated as a within-subject factors. Our results suggest that expert readers never completely "unlearn" the mirror-generalization process and still need to inhibit this heuristic to overcome mirror errors.