The signal fluctuations in Functional Magnetic Resonance Imaging (fMRI) have been proved suitable for investigating brain connectivity. Sequential Monte Carlo methods (“particle filters”) aim at estimating internal states in dynamic systems when only partial and noisy observations are available. Differently from the majority of techniques commonly used for investigating brain connectivity, which assume stationarity, particle filters are designed for stochastic time-varying systems. We present a particle filtering algorithm tailored to fMRI data, whose purpose is to help assessing the presence of causal influences that certain brain areas may exert over others. The algorithm has been validated on a simulated network and applied to real fMRI data.