Oil pipeline leakages, if not properly treated, can result in huge losses. The first step in tackling these leakages is to diagnose their location. This paper employs a data-driven Fault Detection and Isolation (FDI) system not only to detect the occurrence and location of a leakage fault, but also to estimate its severity (size) with extreme accuracy. In the present study, the Golkhari-Binak pipeline, located in southern Iran, is modeled in the OLGA software. The data used to train the data-driven FDI system is acquired by this model. Different leakage scenarios are applied to the pipeline model; then, the corresponding inlet pressure and outlet flow rates are recorded as the training data. The time-domain data are transformed into the wavelet domain; then, the statistical features of the data are extracted from both the wavelet and the time domains. Each of these features are then fed into a Multi-Layer Perceptron Neural Network (MLPNN) which functions as the FDI system. The results show that the system with the wavelet-based statistical features outperforms that of the time-domain based features. The proposed FDI system is also able to diagnose the leakage location and severity with a low False Alarm Rate (FAR) and a high Correct Classification Rate (CCR).