The authors present a multivariate method for selecting the most informative set of correlated variables in a multivariate model. They claim that it is simple and that smaller computing time is needed in comparison with the usual methods, as step-wise, principal components or factor analysis for example. The proposed procedure can be based on the explained variance or on partial correlations. A comparison of the proposed model with respect to usual methods yields that it performs very well in an artificial set of data and in the popular investigation on smoking, used as a touchstone data example by SPSS.