Microdosimetry using Monte Carlo simulation is a suitable technique to describe the stochastic nature of energy deposition by alpha particle at cellular level. Because of its short range, the energy imparted by this particle to the targets is highly non-uniform. Thus, to achieve accurate dosimetric results, the modelling of the geometry should be as realistic as possible. The objectives of the present study were to validate the use of the MCNPX and Geant4 Monte Carlo codes for microdosimetric studies using simple and three-dimensional voxelised geometry and to study their limit of validity in this last case. To that aim, the specific energy (z) deposited in the cell nucleus, the single-hit density of specific energy f(1)(z) and the mean-specific energy were calculated. Results show a good agreement when compared with the literature using simple geometry. The maximum percentage difference found is <6 %. For voxelised phantom, the study of the voxel size highlighted that the shape of the curve f(1)(z) obtained with MCNPX for <1 µm voxel size presents a significant difference with the shape of non-voxelised geometry. When using Geant4, little differences are observed whatever the voxel size is. Below 1 µm, the use of Geant4 is required. However, the calculation time is 10 times higher with Geant4 than MCNPX code in the same conditions.