The paper proposes an approach to detection of Distributed Denial of Service (DDoS) attacks using a modular neural network, which is a series of connected neural networks that solve the problem step by step. The task of DDoS attack detection is decomposed into three interrelated subtasks: detection of anomalous network traffic, detection of DDoS attack traffic and identification of the type of realized DDoS attack, which is especially important due to the tendency of implementing multi-vector DDoS attacks. The results of experimental studies on the quality of performance of the constructed modular neural network confirmed the effectiveness of the proposed approach.