The crystal structure of stoichiometric BaVS3 at room temperature, 100, 60 and 5K has been refined by the total profile analysis based on powder neutron diffraction data. The structural refinement at 100 K is consistent with the model that the driving mechanism for the 240 K transition is the freezing of the dynamical distortion and the static ordering of the V sublattice. The generalised R factor (Hamilton test) indicates that the space group for the phase stable between 240 and 80 K is Cmc21. The structural refinement of the 60 K data shows that the 80 K transition is not accompanied by an additional structural distortion. This means that the transition is not of Peierls type. The electron transfer model proposed by Massenet et al (1980) can explain all the physical properties of the phase stable between 80 and 31 K. The structural refinement of the 5 K data shows that the phase stable between 31 and 5 K has still the Cmc21 space symmetry. No reflections of magnetic origin have been observed at 5 K. This indicates that the magnetic ordering, observed for about half of the V sites at the 31 K transition by inelastic neutron scattering experiments, is not long range. The driving mechanism for the 31 K transition is not yet understood. The disproportionation of the V4+ cations into V3+ and V5+ would explain all the physical properties of the phase stable between 31 and 5 K; however, such a model is unlikely from the crystal chemical point of view. The NMR data of Nishihara and Takano (1981) indicating the presence at 1.3 K of two crystallographically independent V sites are reconciled with the neutron and X-ray diffraction data by assuming that either a fourth transition exists between 5 and 1.3 K or at the 31 K transition the space group changes from Cmc21 to Pmc21.