The inversion of the self-potential data is performed using the modified singular value decomposition for the inverse problem using a linear formulation of the forward problem. The kernel is solved numerically accounting for the topography of the system and the resistivity distribution, which is independently obtained through electrical resistance tomography. A prior constraint based on finite element modelling of ground water flow can also be used to provide a prior source current density model if needed. This self-potential tomography approach is first with a synthetic case study showing how the position of a preferential fluid flow pathway can be retrieved from self-potential and resistivity data and how the seepage velocity can be obtained inside one order of magnitude. This methodology is then applied to a test site corresponding to a portion of an embankment dam along the Rhône River in France. Two self-potential maps (with 1169 and 2076 measurements, respectively) and four resistivity tomograms are used to locate a leak. One self-potential profile and one resistivity profile are used together to perform the 2D inversion of the self-potential data to locate the anomalous leakage at depth and to estimate the flow rate. The depth at which the preferential fluid flow pathway is located, according to self-potential tomography, agrees with an independent geotechnical test using the Perméafor. This demonstrates the usefulness of this methodology to detect preferential water channels inside the body of a dam.