The localization of the sources of power quality disturbances is becoming an important issue in the liberalized energy market. In distribution systems, too many measurement instruments should be used to obtain a complete observation of the electric quantities. Thus, model based state estimation techniques can be implemented to exploit the few real time measurements from the field and all the other available information to estimate the desired quantities. In this paper a novel approach to the estimation of the harmonic sources by means of a Bayesian approach is proposed. The procedures also allows the "quality" of the obtained estimates to be assessed. The theoretical fundamentals of the method are discussed and some preliminary results relevant to a simple distribution network are presented.