This paper presents a method to evaluate residual dependencies between sources estimated by ICA to be used in a hierarchical clustering procedure. As a proximity measure a mutual information-based metric is employed. The properties of each group of components are evaluated at each level of the hierarchical tree by two indices that aim at assessing both cluster tightness and physiological reliability through a template matching process. These two indices are used in three different approaches to find the most suitable combination to explore the hierarchical structure of the clustering. This method is aimed at enhancing late positive event-related brain potentials elicited by emotional picture stimuli. Such critical brain events are produced by presenting a subject with emotionally arousing images with respect to neutral ones. Exploiting the modularity of the spatial distribution of late EEG components, ICA can be employed to separate out their contribution, that is then investigated in an automatic ad objective manner by the clustering procedure.