Twenty-two previously normal children and adolescents who suffered a severe, non-penetrating traumatic brain injury had PET during rehabilitation at a median of 1.5 months after the injury. Outcome was assessed at a median of 25 months after brain injury. 16 subjects had CT or MRI within 24 days of PET and 11 subjects had a second PET at the point of outcome (median 28 months after first PET). The PET score (obtained by adding the score of 15 brain regions: normal metabolism = 1; reduced = 0) was significantly associated with the clinical outcome measure. PET earlier than 12 weeks after head trauma correlated with outcome, but later PET did not. PET scores improved significantly between rehabilitation and outcome for the 11 subjects who had two PETs, but improvement was not associated with improvement in clinical condition. PET score did not add to the amount of variance explained in the last regression model for prediction of outcome when the results of contemporaneous CT/MRI and clinical condition were taken into account. The data suggest that routine PET during rehabilitation is no more useful than contemporaneous CT or MRI for prediction of outcome.