In the non-blind image restoration of the computational optical sectioning microscopy (COSM), the acquisition of accurate point spread function of the system has a major impact on the quality and the stability of image restoration. There are two ways to access the point spread function in general: numerical calculation and physical measurements. Numerical calculation has heavy computational burden and more parameters which always cannot be accurately estimated, hence, it has some limitations in practical application; the point spread function accessed by physical measurements can embody the optical properties of the microscope system most properly, but it has low signal\|to\|noise ratio (SNR), it must be pre-processed before used. A detailed introduction of the principle about how to use the extended Nijboer-Zernike theory (ENZ) to reconstruct the point spread function accessed by physical measurements is given. Experiments proved that this method can rapidly and accurately reconstruct the 3D point spread function of the microscope and raise its SNR.