Abstract: Aimed at the problem of poor detection accuracy and real-time performance of existing malicious domain name detection algorithms for family malicious domain names, a BiLSTM-DAE based malicious domain name detection algorithm is proposed. A Bi-directional long short term memory (BiLSTM) network was used to extract the context sequence features of domain name character, and deep auto-encoder (DAE) was used to extract and classify word formation features of strong characters layer by layer which were similarities within classes and distinctions between classes. The experimental results show that compared with the current mainstream malicious domain name detection algorithm, the algorithm has higher detection accuracy while keeping the detection overhead smaller.