Recent data suggest recovery of language systems but persistent structural abnormalities in the prematurely born. We tested the hypothesis that subjects who were born prematurely develop alternative networks for processing language. Subjects who were born prematurely (n = 22; 600-1250 g birth weight), without neonatal brain injury on neonatal cranial ultrasound, and 26 term control subjects were examined with a functional magnetic resonance imaging (fMRI) semantic association task, the Wechsler Intelligence Scale for Children-III (WISC-III) and the Clinical Evaluation of Language Fundamentals (CELF). In-magnet task accuracy and response times were calculated, and fMRI data were evaluated for the effect of group on blood oxygen level dependent (BOLD) activation, the correlation between task accuracy and activation and the functional connectivity between regions activating to task. Although there were differences in verbal IQ and CELF scores between the preterm (PT) and term control groups, there were no significant differences for either accuracy or response time for the in-magnet task. Both groups activated classic semantic processing areas including the left superior and middle temporal gyri and inferior frontal gyrus, and there was no significant difference in activation patterns between groups. Clear differences between the groups were observed in the correlation between task accuracy and activation to task at P < 0.01, corrected for multiple comparisons. Left inferior frontal gyrus correlated with accuracy only for term controls and left sensory motor areas correlated with accuracy only for PT subjects. Left middle temporal gyri correlated with task accuracy for both groups. Connectivity analyses at P < 0.001 revealed the importance of a circuit between left middle temporal gyri and inferior frontal gyrus for both groups. In addition, the PT subjects evidenced greater connectivity between traditional language areas and sensory motor areas but significantly fewer correlated areas within the frontal lobes when compared to term controls. We conclude that at 12 years of age, children born prematurely and children born at term had no difference in performance on a simple lexical semantic processing task and activated similar areas. Connectivity analyses, however, suggested that PT subjects rely upon different neural pathways for lexical semantic processing when compared to term controls. Plasticity in network connections may provide the substrate for improving language skills in the prematurely born.