The efficiency of the proposed modification of the neural network implementing the principal component analysis (PCA) method is studied. A known neural network – the Hebbian filter- is chosen for the basic method. A test problem that allows varying the complexity of the input vectors is used to generate objects for testing both networks. Three series of experiments were conducted to compare the estimated efficiency of the Hebbian filter and the proposed architecture. The results of the experiments show the proposed modification to have an advantage for all the problems involved.