Introduction: Small animal dosimetry serves as an important link in establishing a relationship between absorbed dose and biological effect during pre-clinical targeted radionuclide therapy. Dosimetric approaches reported to date are based on models aiming at representing the animals used during pre-clinical experiments. However, anatomical variations between models may generate differences in the dosimetric results. Our goal was to assess the impact of the mouse model on the absorbed dose per cumulated activity (S value, in Gy per Bq.s). Methodology: Two datasets were considered. The first one was developed in our laboratory and is a voxel-based model of a 30g female nude mouse. Images were segmented manually to identify more than 30 organs and sub-organs. The second dataset originates from the DIGIMOUSE project. In that model, a 28g normal male nude mouse was used to generate the segmented structures of 9 regions in the head and 12 major organs. A software developed in our laboratory allowed us to read each 3D mouse atlas slice by slice, to crop the mouse volume to remove background air voxels, and to write the geometry description as an input file for the Monte-Carlo code MCNPX using “repeated structure representation”. A linear interpolation scales the voxel size as a function of the total body mass. Results: The comparison of various voxel-based mouse dosimetric models shows that even when scaled to the same total-body mass, models from different mouse breed or gender demonstrate very different organ masses, volumes, and therefore S values. Conclusions: Computation of the S values for pre-clinical studies depends strongly on the definition of the mouse model. Our computational model is a step in the direction of a more realistic description of the geometry in pre-clinical dosimetry.