Multiphoton microscopy (MPM) is an imaging method involving a near infrared range of excitation. However, generating 3D microscopic images in high depths with a satisfying quality is complex due to physical distortions of the beam (optical aberrations and diffraction limit) all long the optical path. The ability of computational approaches to restore reliably deep 3D images rests on the accuracy of the mathematical estimation of optical distortions introduced by the instrument. Instrumental distortions of the beam are characterized by the point spread function (PSF) evolution, especially in the optical axis and whose model can be fitted by a Gaussian shape. In this publication, we are presenting our homemade approach which rests first on the design of an optical phantom constituted by standardized microspheres, immobilized into a 3D volume of 2 mm deep. Then, 3D-volumes of images containing a single object are selected and isolated all along the sample depth using automatic morphological tools. Finally, our computational 3D-Gaussian shape fitting algorithm named FIGARO is applied on each individual PSF. FWHM evolution has thus bean measured for the first time in the 3 dimensions along a 2 mm depth. We have highlighted a highly significant effect of spherical aberrations in our MPM system, showing and increase with a factor 4 of the axial PSF dimensions in the depth while preserving the lateral PSF dimensions.