Patients with phonologic alexia can be trained to read semantically impoverished words (e.g., functors) by pairing them with phonologically-related semantically rich words (e.g, nouns). What mechanisms underlie success in this cognitive re-training approach? Does the mechanism change if the skill is “overlearned”, i.e., practiced beyond criterion? We utilized fMRI pre- and post-treatment, and after overlearning, to assess treatment-related functional reorganization in a patient with phonologic alexia, 2 years post left temporoparietal stroke. Pre-treatment, there were no statistically significant differences in activation profiles across the sets of words. Post-treatment, accuracy on the two trained sets improved. Compared with untrained words, reading trained words recruited larger and more significant clusters of activation in the right hemisphere, including right inferior frontal and inferior parietal cortex. Post-overlearning, with near normal performance on overlearned words, predominant activation shifted to left hemisphere regions, including perilesional activation in superior parietal lobe, when reading overlearned vs. untrained words.