Six stroke patients with clinically significant aphasia were studied 4 weeks and again 12-18 months after their first left hemispheric ictus. The regional cerebral metabolic rate of glucose (rCMRglc) was measured repeatedly by PET at rest and during word repetition, and severity of speech impairment was assessed by a neuropsychologic test battery. The patterns of speech-associated activation of glucose metabolism were related to improvement in language performance as measured by the Token test. Three patients experienced significant recovery from aphasia (Token test: 47 to 3, 45 to 12, and 37 to 5 points, respectively), whereas 3 patients had poor outcome (Token test from 48 to 45, and from 47 to 39 and 24, respectively). Good recovery was related to activation of left hemispheric speech areas surrounding the infarct, especially left superior temporal gyrus. In contrast, the 3 patients with persistent aphasia showed rCMRglc recruitment in right hemispheric regions and were unable to activate left hemispheric speech areas on follow-up. These results indicate that favorable outcome is related to partial sparing of speech areas of the dominant hemisphere that can be (re-) activated. Predominant recruitment of contralateral areas is not efficacious for a considerable recovery from aphasia. It rather indicates unspecific involvement of widespread networks in the effort to perform a complex task.