In this paper we propose a new method for ranking Z-numbers and generalizations. This method is based on the internal structure of the artificial neural network, which suggests that the structure of this network consists of inputs weights and the transfer function linear, nonlinear and sometimes linear and nonlinear. It is shown that the proposed method while possessing the ranking properties for Z-numbers whose components of the limiting part are equal and their confidence interval having the same center of gravity has a more logical ranking than those using the center of gravity. While some of the available methods for Z numbers whose boundaries are equal but not equal to their reliability but have the same focal gravity they rank equally which can not be logical in all cases. Therefore, the proposed method overcomes this problem. In some examples the correctness of the subject is shown. the results are compared with some existing methods.