In bulk forming processes the prediction of ductile fractures and flow defects is a very important task. Generally the ductile fracture criteria are utilized for the estimation of damage accumulated by the material during the deformation. The principal defect of this approach is that each criterion gives a good result for some processes, instead not a good performance is obtained in relation to other processes. Thus, a considerable advantage is obtained by the implementation of a tool able to predict ductile fracture occurrence independently by the particular process analyzed. The aim of this paper is the development of an artificial neural network which can predict the occurrence of fracture with no dependence by a particular criterion. The paper presents the procedure leading to the choice of the most performing network. In particular, the impact of different network  architectures (number of neurons and layers), transfer functions, input vectors, data structures, learning rules was analyzed. The implemented neural network can recognize fracture occurrence in a wide range of bulk forming processes and gives very performing results.