A two-dimensional non-reconstruction adaptive threshold algorithm aiming at infrared small target detection was proposed, for the purpose of decreasing the reconstruction algorithms′ negative influence on target detection′s efficiency and results. Aiming at the two-dimentional measurement model which constructed by Hadamard matrix, the compressive subtract image was analyzed by means of Hadamard′s property in order to decode target′s characteristics in space domain directly. Then the decoded image was detected by utilizing the advanced adaptive threshold method, which avoided the waste of memory space and operation time caused by traditional reconstruction algorithms. Simulation experiment demonstrates that the proposed model can detect the targets on the condition of both single and multiple targets, and has superiorities on detection rate, false alarm rate and operation time than the traditional detection algorithm after reconstruction. It provides a new idea and efficient algorithm for the application of compressive sensing infrared small target detection in engineering.