Two new evaluation of quality approaches for a population of Bayesian networks (BN) are proposed in this paper. The first approach relies on the use of statistical principle with application of well-known evaluation methods. The other bases on quasi essential graph (QEG), an extension of essential graph (EG), that is a presentative graph for all BN of the population. In QEG, each edge is statistically weighted in two parts: (1) undirected part that represents the power of the relationship; (2) arrow part that represents the reliability of the orientation. Results of application to the both simulated and real-world problems show that these proposed approaches are the others helpful solutions for the problem of edge orientation and for the visualization of results of evaluation methods.