Abstract: The article is devoted to the search for the neural network architecture that demonstrates the highest accuracy of assessment of the creditworthiness of borrowers-individuals. There is studied such types of neural network architectures as three-layer perceptron and radial basis function network, as well as the issues of optimal configuration choice. It’s carried out a comparative analysis of the effectiveness of individual neural networks of various architectures and configurations, as well as formed «the committee of experts» from three of the best neural networks. The approach of summarizing in the committee of results of individual models calculations is proposed in the article. Experimental research has confirmed that the combination of several models in the committee allows compensating the possible errors of individual models evaluations. The findings from this study and constructed neural network models may be used by banking institutions and other credit organizations interested in an adequate procedure of assessing the creditworthiness of individuals.