Introduction. Monitoring of the depth of anesthesia during surgery is a complex task. Since electroencephalogram (EEG) signals contain valuable information about processes in the brain, EEG analysis is considered to be one of the most useful methods for study and assessment of the depth of anesthesia in clinical applications. Anesthetics affect the frequency composition of the EEG. EEG of awake persons, as a rule, contains mixed alpha and beta rhythms. Changes in the EEG, caused by the transition from the waking state to the state of deep anesthesia, manifest as a shift of the spectral components of the signal to the lower part of the frequency range. Anesthetics cause a whole range of neurophysiological changes, which cannot be correctly assessed with just one indicator. Objective. In order to describe complex processes during the transition from the waking state to the state of deep anesthesia adequately, it is required to propose a method for assessing the depth of anesthesia, using a comprehensive set of parameters reflecting changes in the EEG signal. The article presents the results of study the possibility of building a regression model based on artificial neural networks (ANN) to determine levels of anesthesia using a set of parameters calculated by EEG. Materials and methods. The authors of the article propose the method for assessing the level of anesthesia, based on the use of neural networks, which input parameters are time and frequency EEG parameters, namely: spectral entropy (SE); burst-suppression ratio (BSR); spectral edge frequency (SEF95) and log power ratio of the spectrum (RBR) for three pairs of frequency ranges. Results. The optimal parameters of ANN were determined, at which the highest level of regression is achieved between the calculated and the verified values of the anesthesia depth indices. Conclusion. In order to create a practical version of the algorithm, it is necessary to investigate further the noise stability of the proposed method and develop a set of algorithmic solutions, which ensure a reliable operation of the algorithm in the presence of noise.