A methodology for pipeline leakage detection, together with isolating the size of the leak and the leakage location, using an evidential reasoning based approach is presented. Leakage in pipes produces two major effects, a change in the difference between inlet flow and outlet flow, and, an average pressure change over time. Each of these changes, although in reality coupled, can be thought of as providing two independent bodies of evidence (typically incomplete and non-specific) which can give hints of the occurance of a leak. Inference using traditional Bayesian analysis involves assumptions in cases of incomplete information and partial ignorance. Evidential reasoning, also called the Dempster-Shafer (DS) theory, has proved very useful for such situations and has the ability to incorporate both aleatory and epistemic uncertainities in the inference mechanism. The bodies of evidence from the changes mentioned above are mapped over a 'frame of discernment' of risk of leakage and subsequently the DS rule of combination is applied to make an inference on the occurence, size and location of a leak. What is described below is a novel methodology as to how a system based on the evidential reasoning approach may be trained to detect and locate leaks in compressible flow in a pipeline. The evidential reasoning based approach is trained off-line with the results encouraging.