The non-invasive multimodal analysis of brain function in the premature babies makes it possible to evaluate the implementation of neural networks involved in various functions such as language. The establishment of neural networks is the result of (1) initial genetic factors that participate in the structural and functional development of networks, (2) endogenous stimuli-independent generators, (3) endogenous stimulus-dependent generators and (4) stimuli-dependant activations. Multimodal analysis, magnetic resonance imaging, optical imaging and high-resolution EEG (HR-EEG) source imaging enable the analysis of these initial premature networks. Thus, as early as 28 weeks gestational age, when the thalamocortical afferents start to connect the cortical plate, when perisylvian structures exhibit certain structural asymmetries, the neural network is able to discriminate voice and phonemes with a neural strategy involving synchronization habituation and mismatch in specific areas, suggesting a strong genetic fingerprint in these initial abilities. Using HR-EEG in premature makes it possible to demonstrate that some endogenous generators such as the theta temporal activities located in the posterior part of the temporal planum are stimuli-independent up to 30-week gestational age and probably participate in the early functional wiring of the perisylvian networks involved in language and communication functions.