ATM system was divided into 3 layers for the purpose of evaluating its information security.An evaluation model was proposed by using a 3-layer artificial neural network（ANN） based on improved BP model.The major information security indicators of ATM system were used as the training samples

which were adapted to find the intrinsic links between the input and output by learning and training process.An experiment was conducted by using the well-trained ANN network to evaluate the security of ATM system.The experimental results show that the proposed ANN evaluation model can indicate the practical running status of ATM system precisely.It is highly adaptive and fault-tolerant.