Brain network modeling is probably the biggest challenge in fMRI data analysis. Higher cognitive processes in fact, rely on complex dynamics of temporally and spatially segregated brain activities. A number of different techniques, mostly derived from paradigmatic hypothesis-driven methods, have been successfully applied for such purpose. This paper instead, presents a new data-driven analysis approach that applies both independent components analysis (ICA) and the Granger causality (GC). The method includes two steps: (1) ICA is used to extract the independent functional activities; (2) the GC is applied to the independent component (IC) most correlated with the stimuli, to indicate its functional relation with other ICs. This new method is applied to the analysis of fMRI study of listening to high-frequency trisyllabic words, non-words and reversed words. As expected, activity was found in the primary and secondary auditory cortices. Additionally, a parieto-frontal network of activations, supported by temporal and causality relationships, was found. This network is modulated by experimental conditions in agreement with the most recent models presented for word perception. The results have confirmed the validity of the proposed method, and seem promising for the detection of cognitive causal relationships in neuroimaging data.