A two-dimensional principal component analysis (2D PCA) method directed at processing digital images is discussed. The method is based on representation of images as a set of rows and columns analyzing these sets. Two methods of realizing the 2D PCA corresponding to the parallel and cascade forms of its realization are presented, and their characteristics are estimated. The application of the 2D PCA method is shown for solving problems of representation and recognition of facial images. The experiments are fulfilled on ORL and FERET bases.