This paper presents a data-driven technique for the localization of leaks in water distribution networks (WDN). The methodology requires hydraulic data, i.e., pressure measurements from a set of sensors installed throughout the network and topological information. Therefore, the hydraulic model of the WDN is not necessary for its operation. The hydraulic state of the complete set of nodes of the network is approximated by means of a graph-based interpolation technique. Then, a set of candidates where the leak can be located is achieved by comparing the computed states for both the leaky and nominal cases. The methodology is applied to a case study based on a real network, providing and discussing several graphical results and key performance indicators.