A method is proposed to determine models of distributed polarizabilities from induction energies calculated quantum chemically for a molecule polarized by a point charge running over a grid of points. Once a polarizability model is chosen, sets of distributed polarizability components recovering exactly the induction energy for a selection of grid points are obtained by a matrix inversion. Distribution functions of the polarizability components are then built and the set of the most probable values for each polarizability component of the model is determined. The quality of a model is assessed by its ability to reproduce molecular polarizabilities and induction energies as accurately as possible. In addition, the distribution function approach provides valuable help to detect ill-defined components and, thus, select the optimal models. Several models of distributed polarizabilities for the water? methanol, and acetonitrile molecules are presented to illustrate the proposed approach.