The reduction of non-technical losses of electrical power (NTL) are currently one of the main challenges faced by electricity utility companies in developing countries. Because it depends in part on factors not manageable by the utilities, in Brazil is passed on to consumers through electricity tariffs, following the limits defined by the National Electric Energy Agency (ANEEL). Despite the notable progress made by ANEEL, the currently used methodology by this Agency to define these limits has a number of limitations, which have impose the adoption of ad hoc measures that end up resulting in loss targets that are inconsistent with the reality of the utilities. In the last three decades, a several mathematical models of Efficiency Analysis have been developing in order to measure the relative efficiency of groups of agents. However, the adoption of these models by Regulator Bodies around the world has been limited to subjects related to operating costs and investment. This dissertation aims to fulfill this gap, proposing an alternative model for the definition of NTL regulatory targets from a particular family of Efficiency Analysis models - the Stochastic Frontier Analysis (SFA) models. The proposed model has been applied to a data panel containing annual observations of 62 distribution concessionaires in the Brazilian electrical system from 2007 to 2017, resulting in a set of NTL targets more adherent to the real PNT of Brazilian electricity utility distributors.