In this paper, a detailed study of the electron transport in MCNP is performed, separating the effects of the energy binning technique on the energy loss rate, the scattering angles, and the sub-step length as a function of energy. As this problem is already well known, in this paper we focus on the explanation as to why the default mode of MCNP can lead to large deviations. The resolution dependence was investigated as well. An error in the MCNP code in the energy binning technique in the default mode (DBCN 18 card = 0) was revealed, more specific in the updating of cross sections when a sub-step is performed corresponding to a high-energy loss. This updating error is not present in the ITS mode (DBCN 18 card = 1) and leads to a systematically lower dose deposition rate in the default mode. The effect is present for all energies studied (0.5-10 MeV) and depends on the geometrical resolution of the scoring regions and the energy grid resolution. The effect of the energy binning technique is of the same order of that of the updating error for energies below 2 MeV, and becomes less important for higher energies. For a 1 MeV point source surrounded by homogeneous water, the deviation of the default MCNP results at short distances attains 9% and remains approximately the same for all energies. This effect could be corrected by removing the completion of an energy step each time an electron changes from an energy bin during a sub-step. Another solution consists of performing all calculations in the ITS mode. Another problem is the resolution dependence, even in the ITS mode. The higher the resolution is chosen (the smaller the scoring regions) the faster the energy is deposited along the electron track. It is proven that this is caused by starting a new energy step when crossing a surface. The resolution effect should be investigated for every specific case when calculating dose distributions around beta sources. The resolution should not be higher than 0.85*(1-EFAC)*CSDA, where EFAC is the energy loss per energy step and CSDA a continuous slowing down approximation range. This effect could as well be removed by determining the cross sections for energy loss and multiple scattering at the average energy of an energy step and by sampling the cross sections for each sub-step. Overall, we conclude that MCNP cannot be used without a caution due to possible errors in the electron transport. When care is taken, it is possible to obtain correct results that are in agreement with other Monte Carlo codes.