If we predict the host range of new or mutant influenza virus in advance, we are able to measure against pandemic human influenza immediately after the new virus emerges somewhere. Influenza viral hemagglutinin(HA)-sialoside receptor interaction is a target event for in silico chemical prediction studies about the virus host range determination. We theoretically studied avian and human influenza A virus HA H3 subtype complexed with avian or human type receptor Neu5Acalpha(2-3 or 2-6)Gal analogues by ab initio fragment molecular orbital (FMO) method at the second order M?ller-Plesset (MP2)/6-31G level, which can evaluate correctly not only electrostatic interactions but also lipophilic interactions based on van der Waals dispersion force. Avian H3 bound to avian alpha2-3 11.4 kcal/mol stronger than to human alpha2-6 in the model complexes with taking account of intermolecular lipophilic interaction. A substitution at the position 226 between Gln(avian) and Leu(human) on influenza H3 HA1 has altered its virus host range between avian and human. In the ab initio FMO studies, binding energy of avian Gln226Leu H3-human alpha2-6 was quite similar to that in the human H3-human alpha2-6 complex with amino acid sequence differences at nine positions in the models. This similarity indicates that avian Gln226Leu H3 virus can infect human with the same level as human H3 virus. Opposite mutation Leu226Gln in the human H3 gave the moderate binding energies to avian alpha2-3 with similarity to avian H3-alpha2-3 complex that supported our previous virus-sialoside binding assay. Ab initio FMO studies have revealed the relationship between influenza H3 virus host range and H3-alpha(2-3 or 2-6) receptors binding. Our theoretical approach may predict the infectious level of new viruses and point out some unknown dangerous mutation positions on HA in advance.