The ultimate goal of projection methods is to search “interesting” projections in a low‐dimensional subspace that can uncover the natural structure of the data. The aim of this work is to compare the ability of projection pursuit (PP) and principal component analysis (PCA) in dimension reduction. For this purpose to be achieved, the scores of PP and PCA, by a different number of factors, were used as inputs of radial basis function (RBF) neural network. RBF neural network was used as a nonlinear regression method in a quantitative structure‐retention relationships study of 209 polychlorinated biphenyls (PCBs). The dependent variable was the highresolution gas chromatographic relative retention times of PCBs on 18 different stationary phases, and independent variables were solvatochromic solute descriptors. The results demonstrate that the dimension reduction ability of the PP is better than that of the PCA for both single and full column retention models.
