Standard macroscopic methods used to assess the dose in nuclear medicine are limited to cases of homogeneous radionuclide distributions and provide dose estimations at the organ level. In a few applications, like radioimmunotherapy, the mean dose to an organ is not suitable to explain clinical observations, and knowledge of the dose at the tissular level is mandatory. Therefore, one must determine how particles lose their energy and what is the best way to represent tissues. The Monte Carlo method is appropriate to solve the problem of particle transport, but the question of the geometric representation of biology remains. In this paper, we describe a software (CLUSTER3D) that is able to build randomly biologically representative sphere cluster geometries using a statistical description of tissues. These geometries are then used by our Monte Carlo code called DOSE3D to perform particle transport. First results obtained on thyroid models highlight the need of cellular and tissular data to take into account actual radionuclide distributions in tissues. The flexibility and reliability of the method makes it a useful tool to study the energy deposition at various cellular and tissular levels in any configuration.