This paper presents computationally efficient artificial neural network technique for assessing the security of the power system against line outages. Performance index (PI), which accounts for various line limit violations per contingency, is defined. The basic purpose of ANN is to assess the severity of line outages in terms of PI, based on training examples from off-line analysis. The selection of input signals for ANN is influenced by the operating state of the system and the contingency in question which determines the extent of line power limit violations. In an attempt to attain perfection in PI prediction, suitable architecture and topology for the network is investigated. To expedite learning process, saturating linear coupled neuron model (sl-CONE) is also tried out. The effectiveness of proposed technique is demonstrated on 5-bus (7-line) and IEEE 14-bus (21-line) test systems. Computation efficiency of the method makes it potential candidate for inclusion in on-line comprehensive security analysis package. © 2002 Elsevier Science Ltd. All rights reserved.