In three-dimensional fluorescence microscopy the point spread function (PSF) changes with depth, inducing errors in the restored images when these variations are neglected during the deconvolution of thick specimens. Some deconvolution algorithms have been developed to take the depth variations of the PSF into consideration. For these algorithms, the accuracy of the estimated structures depends on the knowledge of the PSF at various depths. We propose an alternative to measuring all required PSFs at different depths. The needed PSFs are interpolated from a limited measured PSF set using a method based on Zernike moments. The proposed method offers the possibility to obtain an accurate PSF interpolation at different depths using only a few measured ones.