In paper proposed method for detecting and classifying cyberbullying in text content using neural networks, which allows for detecting both the general level of cyberbullying presence in a message and performing multi-label classification by individual types of cyberbullying. The method allows for separately determining the manifestation levels of age, religious, ethnic, gender, and other types of cyberbullying. To research the method effectiveness, appropriate software was developed. For binary classification, the results were Accuracy 96%, Precision 96%, Recall 95.9% and F1 95.7% using BiLSTM neural network model. For multi-label classification, the results were Accuracy 94%, Precision 93%, Recall 93% and F1 93% using BERT neural network model, and Accuracy 89.8%, Precision 92.1%, Recall 95.6% and F1 93.8% using L-BFGS Maximum-EntropyMulti neural network model. The results obtained allow for more effective cyberbullying detection in text messages, as well as for determining the prevalence rate of each type of cyberbullying. The developed method for detecting cyberbullying complies with goals SDG3 (good health and well-being), SDG4 (quality education) and SDG16 (peace, justice, and strong institutions), and can be implemented in social media platforms, educational platforms, in social platforms moderation systems to provide psychological assistance to victims or witnesses of cyberbullying.