We consider the description of the problem of identifying the parameters of the input-output representation of the system, based on the second-order entropy in the notion of Rényi's identification error. In this case, the identification criterion is maximized for all possible distributions and then minimized for the vector of system parameters. A heterogeneity measure is then constructed using the Rényi quadratic entropy, which allows the input variables (structural identification as an initial step) to be selected in a suitable way, justified by the properties of the proposed heterogeneity measure, for models subjected to further parametric identification. In this case, the properties of the proposed measure of dependence, namely the consistent normalization of the measure of dependence, are very important.