In previous studies we demonstrated that a relatively large diversity of stress response patterns (acid, osmotic, oxidative, heat) exists among Streptococcus thermophilus strains. Changes in protein expression, evaluated by SDS-PAGE in 4 wild strains (CNBL7035, TH681, Y3, Sfi39) and in three Sfi39 mutants in which hrcA, ctsR and rr01 genes were inactivated showed that significant variations of proteins involved in general stress response (GSR) occur as a function of growth phase, adaptation and inactivation of stress response regulators. In this work we re-evaluate the previous results comparing two unsupervised (Hierarchical Cluster Analysis, HCA, and Principal Component Analysis, PCA) and one supervised (Partial Least Square Regression, PLSR) statistical techniques for the ability to extract information from SDS-PAGE patterns of wild type and mutant strains of S. thermophilus and to uncover relationships between protein patterns and stress tolerance. HCA and PCA are two purely descriptive techniques. The HCA showed that SDS-PAGE is an efficient tool to differentiate strains but did not shed any light on the relationships between band intensity and strain, growth phase or adaptation treatment. PCA helped to identify group of bands which covaried with the stress input factors but also not allow to find a relationship between protein expression and stress tolerance. The PLS regression, even with the limitations due to the data set used in this study, appears as an extremely promising tools for the identification of complex relationships between design and response variables in the analysis of SDS-PAGE patterns of whole cell proteins.