Abstract: Malicious domain name detection is of great significance to prevent botnet and other network attacks. This paper proposes a malicious domain name detection method called CNN-BiGRU-Focal. Convolutional neural network and bidirectional gated cyclic unit network were used for feature fusion learning, and an improved focal loss function was introduced to solve the problem of data imbalance. Compared with LSTM, CNN, GRU and ATT-CNN-BiLSTM method, the detection accuracy of the proposed method is improved by 1.43, 2.89, 1.27 and 2.43 percentage points in multi-classification experiments, and 0.19, 0.12, 1.41 and 0.3 percentage points in binary classification experiments. Experiments show that CNN-BiGRU-Focal method has better performance in the detection of malicious domain names.