Landslide zoning by means of statistical analyses may be carried out in two successive steps, respectively producing: a landslide susceptibility computational map; a susceptibility map for zoning purposes. The procedure proposed herein is relevant for the first step of the analysis, as it allows to select the optimal territorial variables (i.e. independent variables of the statistical analysis) derived from significant thematic maps available over the territory to be zoned. To this aim, different classification schemes and focal statistics techniques are adopted. The selection process of the variables is based on a series of bivariate analyses between a set of different independent variables derived from the same significant thematic map and an inventory of landslides (i.e. dependent variable of the bivariate analysis). At the end of the process, one optimal independent variable for each significant thematic map is selected on the basis of performance indicators taking into account: the contribution to the success of the analysis; the discriminant capability of the classification; the accuracy of the analysis. The proposed method is applied to a case study of shallow landslides in fine-grained soils, in a test area located in the province of Catanzaro of the Calabria region (southern Italy).