Relational Bayesian networks are a highly expressive representation language for probability distributions on relational structures. Distinguishing features of this framework are syntactic simplicity (the language is defined by only four elementary syntactic constructs loosely corresponding to the constructs of predicate logic) and semantic clarity. In this paper I give an overview over relational Bayesian networks. First syntax, semantics and inference methods for elementary queries are recapitulated, then I will highlight some of the more challenging topics that arise when one asks non-elementary queries about probabilities for infinite structures, or limit probabilities for finite structures of increasing size. Finally, I will briefly discuss the question of learning relational Bayesian networks from data.