(eng): We compare two approaches to the construction of belief networks in the framework of evidence theory. We show that belief networks reconstructed from multidimensional models based on recently introduced operator of composition of basic assignments have similar structural properties to Bayesian networks. On the other hand, it is not valid for previously proposed directed evidential networks by Ben Yaghlane at al. The difference between the structural properties of models based on these two approaches is illustrated by a simple example.