Two patients (G01, J02) with chronic nonfluent aphasia and sentence production deficits received syntactic mapping treatment to improve sentence production. The patients had dramatically different outcomes in that improved syntax production generalized to nontreatment tasks for G01, but not for JO2. To learn how treatment influenced the neural substrates for syntax production, both patients underwent pre- and posttreatment functional magnetic resonance imaging (fMRI) of sentence generation. G01 showed more robust activity posttreatment than pretreatment in Broca's area; ventral temporal activity decreased slightly from pre- to posttreatment. Comparison of J02's pretreatment and posttreatment images revealed little change, although activity was more diffuse pre- than posttreatment. Findings suggest that for G01, rehabilitation led to engagement of an area (Broca's area) used minimally during the pretreatment scan, whereas for J02, rehabilitation may have led to more efficient use of areas already involved in sentence generation during the pretreatment scan. fMRI findings are discussed in the context of sentence-production outcome and generalization.