The study of time-varying connectivity networks is a young but growing field of research in functional MRI, where dynamic Bayesian networks (DBNs) should play an important role for many reasons. In this paper, Product Hidden Markov Models (PHMMs), an instance of DBNs, are introduced to capture the dynamic functional connectivity (DFC) of spontaneous co-activation maps (SAMs), including resting state networks (RSNs), at the subject level. The abilities of PHMMs to learn dependencies between interacting processes are illustrated to compare and analyze inter-sessions DFCs of healthy subjects taking medication (methylphenidate) vs a placebo. PHMMs are presented as a novel methodology for characterization of and knowledge extraction from the DFC.