Recent brain imaging studies suggest that semantic processing of words and images may share a common neural network, although modality-specific activation can also be observed. Other studies using event-related potentials (ERPs) report that brain responses to words and images may already differ at approximately 150 ms following stimulus presentation. The question thus remains, which differences are due to perceptual categorization processes and which differences are due to the semantic ones? Using ERP recordings and spatio-temporal source localization analysis, we investigated the dynamics of brain activation during a recognition task. The stimuli consisted of a randomized set of verbal (words vs. non-words) and pictorial items (line drawings of objects vs. scrambled drawings). After each stimulus, subjects had to decide whether it corresponds to a recognizable word or objects. ERP map series were first analyzed in terms of segments of quasi-stable map topography using a cluster analysis. This showed that verbal and pictorial stimuli elicited different field patterns in two time segments between approximately 190-400 ms. Before and after this period, map patterns were similar between verbal and pictorial conditions indicating that the same brain structures were engaged during the early and late steps of processing. Source localization analysis of map segments corresponding to the P100 and the N150 components first showed activation of posterior bilateral regions and then of left temporo-posterior areas. During the period differentiating conditions, other patterns of activation, involving mainly left anterior and posterior regions for words and bilateral posterior regions for images, were observed. These findings suggest that, while sharing an initial common network, recognition of verbal and pictorial stimuli subsequently engage different brain regions during time periods generally allocated to the semantic processing of stimuli.