There is an ongoing debate in cognitive neuroscience about the time course and the functional independence of the different processes involved in encoding written language material. New data indicate very fast and highly parallel language analysis networks in the brain. Here we demonstrate a methodological approach to study the temporal dynamics of this network by searching for time periods where different task demands emphasize different aspects of the network. Multi-channel event related potentials (ERPs) were recorded during a semantic and a phonological reading task from 14 healthy subjects. Signals were analyzed exclusively on the basis of the spatial configuration of the electric potential distributions (ERP maps), since differences in these spatial patterns directly reflect changes in the configuration of the active sources in the brain. This analysis did not reveal any differences of the evoked brain electric fields between the two tasks up to 280 ms post-stimulus. The ERP maps then differed for a brief period between 280 and 380 ms, before they were similar again. The analysis of the maps using a global linear localization procedure revealed a network of areas, active in both tasks, that mainly involved the left postero-temporal and left antero-temporal regions. The left posterior activation was found already around 100 ms post-stimulus, indicating that language-specific functions appear early in time. We therefore conclude that phonological and semantic processing are essentially performed in both tasks and that only late decision-related processes influence the relative strength of activity of the different modules in the complex language network.