The first part of the presentation is a brief survey of the neural network models and their use in power systems, based on an analysis of more than 200 publications. The second part discusses the applications in static security assessment, with supervised and unsupervised neural network models. An example of static security assessment with the Kohonen classifier is developed: the Kohonen network is trained with vectors representing simulated states of the power system, with or without contingencies. The Kohonen map is used to assess unknown power system states in real time. The two approaches, supervised and unsupervised, are compared and finally, the perspective of industrial implementation is evaluated