Abstract: An on-line leak-detection method based on pressure time series of oil pipeline is proposed for leakage. Reconstructing the phase-space of pressure time series of oil pipeline, the numbers of embedding dimensions and embedding computation delays are given as 5 and 4, respectively. Taking the 5-D reconstructing vectors as the input to the neural network model, the proposed method is used to train the network off-line so as to obtain the initial parameters and, then, train the model on-line to implement the on-line modification of the weighted values in the network model, thus realizing the real-time detection of leak. The datasets measured in simulation revealed that the method proposed is available to detect pressure faults efficiently.