Abstract: ‍ ‍In recent years， international terrorist attacks have occurred frequently， making the security situation extremely severe. Therefore， there is an urgent need to introduce new security screening methods to prevent and respond to major security incidents. Consequently， millimeter wave holographic three-dimensional （3D） imaging has been widely used in the security-screening field because of its security， penetration， and high-resolution imaging advantages. Because human body security screening requires accurate detection of targets within a short period， it has the characteristics of real-time， requiring quick judgments and decisions in real-time； therefore， it has higher requirements for imaging precision. In this paper， a high precision millimeter wave cylindrical aperture holographic imaging algorithm is proposed， and the sampling criteria for 3D imaging of cylindrical aperture are presented. First， the imaging region is divided into imaging cylindrical surfaces with different radii， and the imaging reference cylindrical surfaces are set according to the imaging accuracy and phase error control range to simplify the process of repeated calculation of the matching filter function. Then， the echo signal is transformed into the 3D wavenumber domain and multiplied with the matching filter function. Next， the results of matched filtering are integrated along the range direction， and the scattering information of the target on the corresponding radius cylindrical surface is obtained by focusing. Finally， the target on the cylindrical surface with a different radius is focused to obtain all the scattering information of the target， and then， the complete 3-D image reconstruction of the entire observation area is achieved. Moreover， the optimal allocation strategy of reference surface and the algorithm imaging complexity analysis are presented. The effect of phase error on imaging quality is analyzed in detail through point target simulation and compared with existing methods. Additionally， imaging experiments on human body models under the same system parameters are conducted. The simulation and experiment results show that the proposed method is effective， accurate， and correct.