Abstract: The existing approaches of anomaly diagnose for database servers need a large amount of computation and consume a lot of system resources. This paper incorporated the dendritic cells algorithm(DCA), which was based on danger theory, into database servers' anomaly detection. Firstly, it analyzed runtime characteristics of database servers. Secondly, to evaluate server's characteristics, it built multidimensional metrics which were normalized between 1 and 10. Finally, it performed DCA on data that we collected from database servers. An empirical analysis on the dataset revealed that our approach performed well on improving detection accuracy and reducing false positive rate.