Graphical User Interfaces (GUI) software has characteristics different from traditional software. The oracle for GUI software testing must validate the correctness of the GUI. An automated oracle based on multi-weighted Neural Networks (NN) is proposed in this paper to validate the GUI from users` viewpoint. In this approach the multi-weighted NN is used to learn the topological information in the feature space for the expected images of the graphical interface. The topological information is then used to verify the correctness of the GUI. By this method, trivial difference in the graphical interfaces can be ignored and GUI be automatically tested in the manner of human being. Experimental results show the method is of potential application in automated GUI testing.