The mismatch negativity (MMN), a well-known index of automatic acoustic change detection, was recently found to be a sensitive indicator of long-term memory traces for native language sounds. When comparing MMNs to words and meaningless pseudowords of Finnish and English using EEG and MEG, we found larger MMN amplitudes for words than meaningless items. This effect was independent of the grammatical status of the standard stimulus. We have also demonstrated that, using MMN, it is possible to register differences in the brain response to individual words, suggesting that the cortical memory networks of individual lexical items can be revealed by the MMN. In more recent studies, we found additional evidence that the mismatch negativity reflects automatic syntactic processing commencing as early as ~100ms after the relevant information becomes available in acoustic input. The findings will be discussed using a distributed neuronal network approach. In summary, neurophysiological imaging of the mismatch negativity response provides a unique opportunity to see subtle spatio-temporal dynamics of the language processing in the human cortex.