The development of complex and highly realistic virtual simulators involves the collection and analysis of a large amount of data about the process of human interaction with the virtual environment. One of the key sources of data in the field of professional training is the physical, including muscular activity of the user, on the basis of which conclusions can be drawn about the person’s condition and his preparation. To generate a set of such data, various electromyographs are used. The work discusses an approach to collecting and processing electromyography data for the subsequent solution of the problem of classifying user movements. Classification is performed using various machine learning methods, including dense and convolutional neural networks. As a result of the experiment on classification of 6 movements, an accuracy of 95% was obtained on a convolutional neural network. The experiment also analyzed the influence (correlation) of different electromyography channels on a specific movement. The results obtained can be used in professional training complexes as part of biofeedback or for collecting and processing information about the user’s muscle activity.