Knowledge of spatiotemporal patterns of language network changes may help in predicting outcome in aphasic stroke patients. Here we assessed language function and performed functional MRI four times during one year to measure language network activation and cerebrovascular reactivity (with breath-holding) in twelve left-hemispheric stroke patients, of whom two dropped out before the final measurement, and eight age-matched controls. Language outcome was related to increase of activation in left and right posterior inferior temporal gyrus over the first year, while activation increase in right inferior frontal gyrus was inversely correlated to language recovery. Outcome prediction improved by addition of early language-induced activation of the left posterior inferior temporal gyrus to a regression model with baseline language performance as first predictor. Variations in language-induced activation in right inferior frontal gyrus were primarily related to differences in vascular reactivity. Furthermore, several language-activation changes could not be linked to alterations in language proficiency nor vascular reactivity, and were assumed to be caused by unspecified intersession variability. In conclusion, early functional neuroimaging improves outcome prediction of aphasia after stroke. Controlling for cerebrovascular reactivity and unspecified intersession variability may result in more accurate assessment of the relationship between activation pattern shifts and function after stroke.