The ability of functional magnetic resonance imaging (FMRI) to noninvasively measure fluctuations in brain activity in the absence of an applied stimulus offers the possibility of discerning functional networks in the resting state of the brain. However, the reconstruction of brain networks from these signal fluctuations poses a significant challenge because they are generally nonlinear and nongaussian and can overlap in both their spatial and temporal extent. Moreover, because there is no explicit input stimulus, there is no signal model with which to compare the brain responses. A variety of techniques have been devised to address this problem, but the predominant approaches are based on the presupposition of statistical properties of complex brain signal parameters, which are unprovable but facilitate the analysis. In this article, we address this problem with a new method, entropy field decomposition, for estimating structure within spatiotemporal data. This method is based on a general information field-theoretic formulation of Bayesian probability theory incorporating prior coupling information that allows the enumeration of the most probable parameter configurations without the need for unjustified statistical assumptions. This approach facilitates the construction of brain activation modes directly from the spatial-temporal correlation structure of the data. These modes and their associated spatial-temporal correlation structure can then be used to generate space-time activity probability trajectories, called functional connectivity pathways, which provide a characterization of functional brain networks.