Bayesian networks for large and complex domains are difficult to construct and maintain. For example modifying a small network fragment in a repetitive structure might bevery time consuming. Top-down modelling may simplify the construction of large Bayesian networks, but methods (partly) supporting top-down modelling have only recently been introduced and tools do not exist. In this paper, we try to take a top-down approach to constructing Bayesian networks by using existing object oriented methods. We change these where they fail to support top-down modeling. This provides a new framework that allows top-down methodologies for the construction of Bayesian networks, provides an efficient class hierarchy and a compact way of specifying and representing temporal Bayesian networks. Furthermore, a conceptual simplification is achieved.