In order to improve the object identification capacity of our passive human terahertz imaging system, a passive terahertz image reconstruction algorithm is put forward using deconvolution method. First the main reasons of image degradation are analyzed according to the basic imaging principle of the system. Then the original terahertz images are de-noised as preprocessing course. Afterwards the images are reconstructed by deconvolution method using Gaussian point spread function, so the image resolution is improved. Their grayscales are corrected. The original human terahertz images are processed and resulting images with higher definition and better object identification capability are obtained. Experimental results show that the proposed algorithm is able to effectively improve the image quality of our passive human terahertz imaging system. It can help the observers detect the contrabands hidden in the clothes of imaging subjects more quickly and accurately, which consequently strengthens the practicability of our imaging system.