Bayesian networks can be used to express the probability distributions on the graph pattern, and provide a convenient method to express the probability information, and can express cause relations, but do not limit the cause relations. Bayesian networks have strong reasoning ability in solving nondeterministic problems, and attract more and more attentions from lots of researchers. Orienting arcs of the Bayesian network are part of the learning network, which means to determine the directions of the edges after getting the dependence structure of a Bayesian network.We introduce an improved algorithm about orienting edges which combines the advantages of some other algorithms. Experimental results show that this algorithm can effectively orient the edges of Bayesian networks.