The works deal with multivariate classifications (by supervised and unsupervised pattern recognition techniques) and factor analyses on various kinds of data. The fundamental aim of these studies lies in assessing the actual limits of effectiveness of the techniques used, by testing them under unusual conditions and in the presence of various data transformations and statistical distributions of  variables. The subjects are presented in the following order. (1) Definition of the most suitable kind of data (standardized, transformed, normally distributed) for a realistic factor analysis. (2) Factor analysis of thermodynamic data for solution equilibria, in order to point out and to assess quantitatively the causes giving origin to the processes studied. (3) Joint use of factor analysis and supervised  pattern recognition in characterizing ancient pictorial binders. (4) Definition of the most suitable treatment of responses below the detection limit in multivariate analysis of data. (5) Comparison of various data transformations in multivariate classification of samples. (6) Possible solutions of the problem concerning the presence of ‘radial’ data in analytical chemistry.