A regression strategy based on k nearest neighbors (k-NN) to estimate the fault distance in radial power systems is proposed. This fault location approach uses measurements of the fundamental components of voltage and current measured at the power substation. In addition, the approach is not constrained by the power system modeling and it is easily adaptable to the special characteristics of radial systems. The proposed fault locator is tested in a power distribution system and the obtained mean error is lower than 3%, by considering all fault types, several faulted nodes and fault resistances.