This study describes a treatment project, carried out with two anomic subjects. RBO and GMA failed to name pictures correctly as a consequence of damage to phonological lexical forms; their ability to process word meaning was unimpaired. Words that were consistently comprehended correctly, but produced incorrectly by each subject, were identified. Some words were treated, whereas some served as the control set. A significant improvement was observed in both subjects. As predicted by the model of lexical-semantic processing used as the theoretical background for the study, improvement was restricted to treated items and did not generalize to untreated words, not even to words that were semantically related to those administered during treatment. Improvement was long-lasting, as shown by the fact that 17 months post-therapy GMA's performance on treated words was still significantly better than before treatment. These results are discussed in relation to the claim that cognitive models can be profitably used in the treatment of language disorders.