The S1 and S2 subunits of the spike glycoprotein of SARS-CoV coronavirus have been modelled, even though the corresponding amino acid sequences were not suitable for tertiary structure predictions with conventional homology and/or threading procedures. An indirect search for a protein structure to be used as a template for 3D modelling has been performed on the basis of the similarity, generally exhibited by coronaviruses, of their genomic organisation. The crystal structure of Clostridium botulinum neurotoxin B appeared to be structurally adaptable to human and canine coronavirus spike protein sequences and it was successfully used to model the two subunits of SARS coronavirus spike glycoprotein. In facts, all the possible cystine bridges, ten and four disulphide bridges respectively in the S1 and S2 subunits, can be formed in the two proposed models, since always the corresponding cysteine C atoms are found at a maximum distance of 10 Å. Furthermore, only three out of the total fourteen predicted glycosilation sites of S1 exhibit low surface accessibility and all of the five possible glycosilation sites result to be exposed in S2. The overall shape and the surface hydrophobicity of the two subunits in the obtained models suggest the localisation of the most relevant regions for their activity.