Abstract: In this article the authors offer a method for constructing logical operations to analyze and correct the results of the operation of sigma-pi neural networks designed to solve recognition problems. The aim of the research is to reveal the logical structure of implicit regularities formed as a result of training the neural network. The method proposed by the authors restores the training sample based on the values of the sigma-pi weighting coefficients of the neuron, analyzes the relationships of this structure and allows to detect implicit regularities, which contributes to the increase of the adaptive properties of the sigma-pi neuron. To solve this problem, the authors perform a logical-algebraic analysis of the subject area within the framework of which the cigma-pi of a neuron is trained, a logical decision function is constructed, its properties and applicability to the correction of the work of a neuron are investigated. It is widely known that the combined approach to the organization of the recognition algorithms increases their effectiveness. The authors argue that the combination of the neural network approach and the use of logical correctors allows, in cases of an incorrect response, to indicate the object closest to the requested attributes from the sample on which the sigma-pi neuron was trained. This significantly improves the quality of the automated solution of intellectual problems, i.e. ensuring the accuracy of achieving the right solution by using the most effective systems for analyzing the original data and developing more accurate methods for their processing.