Let $G = (V,E)$ be a directed acyclic graph and $X = \{X_v\}_{v \in V}$ a set of random variables. $(G,X)$ is a Bayesian network if and only if it satisfies the Local Markov property.