This paper presents a model validation method based on neural network by analyzing, in detail, the subsistent problem of model verification method in application. In this method, by using the powerful ability of nonlinearity mapping of neural network and learning the key properties of the behaviors of various models, the real system behaviors are classified as coming from one of the models, thereby evaluating the credibility of the models. In the concrete process, input and output of models are used as the training set to train neural network firstly, then the matching degree of models and real system is ascertained by the sendout value (probability vector) of the trained neural network when the actual system acts on the trained neural network. Simulation results of the final image target recognition models lastly further show the feasibility and validity of this method. Therefore, this model verification method based on neural-network can be used to verify the modeling nicety degree of complex uncertainty system.