Substantial neurobiological data indicate that the dominant cortical region for printed-word recognition shifts from a temporo-parietal (dorsal) to an occipito-temporal (ventral) locus with increasing recognition experience. The circuits also have different characteristic speeds of response and word preferences. Previous evidence suggested that grain-size increased with word repetition. In the present experiment, we examined morpheme boundaries imposing a limitation on steadily increasing grain-size within a word. We manipulated the type font within a word so that it either emphasized or disguised the word's morphemic structure. The results showed that, even after several repetitions, morphemic structure had an effect on word recognition, although eventually it did become nonsignificant. Thus, the spread of grain-size with repetition does not appear to cross morpheme boundaries easily. These results suggest that skilled, experienced, word recognition may achieve its speed, substantially, by unitizing the word's letters within but not across its morphological units.