We have investigated the possibility of predicting the distribution of ions of different charge during electrospray ionization of peptides in mass spectrometric experiments using neural networks. Three independent data sets obtained on the same equipment and deposited in ProteomeXchange (PXD032141, PXD051750, PXD019263) were used as training and test samples. A set of fractional values for 1+ to 5+ ions was calculated as predicted values for each of the newly identified peptides. Four different sets of peptide descriptions were used as independent variables, including both the spectrum of amino acid residues and the physicochemical properties of the amino acid residues. Sixty-four variants of neural networks were analyzed, varying the input description, number and type of layers, activation and loss functions. The coefficient of determination and a set of Euclidean, Sørensen, Chebyshev, and Cosine metrics were considered as measures of prediction quality. For the best selected variants, the error did not exceed 10% in 80% of the cases. This accuracy may be sufficient for a preliminary estimation of the probability of detecting a peptide ion of a given charge.