The advent of functional Magnetic Resonance Imaging (fMRI) has significantly improved the knowledge about the neural correlates of perceptual and cognitive processes. The aim of this thesis is to discuss the characteristics of different approaches for fMRI data analysis, from the conventional mass univariate analysis (General Linear Model - GLM), to the multivariate analysis (i.e., data-driven and pattern based methods), and propose a novel, advanced method (Functional ANOVA Models of Gaussian Kernels - FAM-GK) for the analysis of fMRI data acquired in the context of fast event-related experiments. FAM-GK is an embedded method for voxel selection and is able to capture the nonlinear spatio-temporal dynamics of the BOLD signals by performing...