A method of the concealed object segmentation based on maximum two-dimensional entropy for passive terahertz security is proposed. The method firstly employs a filter bank to reduce image noise. A self-generated detection region algorithm is designed, which can automatically cover the key detection area. The concept of two-dimensional entropy is introduced to implement the concealed object segmentation. Evaluation and comparison experiments are conducted in 0.2 THz band passive images, demonstrating that the method has a good segmentation performance and real-time performance. It may have an important application in the automatic detection for terahertz security.