G3(QCI) model chemistry,which eliminates the uncertainties of basis sets and theoretical methods extrapolation approximations,is defined via eq.: E0[G3(QCI)]=E[QCISD(T,FC)/G3large]＋Δ E(full)＋ E(SO)＋ E(ZPE)＋ E(HLC) where the electronic energy E[QCISD(T,FC)/G3large] is directly calculated at QCISD(T,FC)/G3large level,the core electron correlation energyΔ E(full) is calculated at MP2/6-31G(d) (denoted as fu1) or MP2/6-311G(d,p) (denoted as fu2) level and the spin-orbit coupling E(SO) for atomic species is from the experiments as what has been done in the G3[2] theory.Four varieties of G2(QCI) model chemistry theoretical procedure,namely G3(qCI/fu1),G3(qCI/fu2),G3(QCI/fu1)//B3PW91;G3(QCI/fu2)//B3PW91,are proposed and tested with the G2-1 test set.The E(HLC)'s have been defined and optimized in the same manner as in G2 theory,which are -3.024na-5.397nβ on molecules and -2.085na-5.636nβ on atomic species for G3(QCI/fi1),-3.085na-5.478nβ on molecules and -1.885na-5.466nβ on atomic species for G3(QCI/fu2),-2.495na-5.478nβ on molecules and -1.70na-5.584nβ on atomic species for G3(QCI/ful)//B3PW91 and -2.670na-5.380nβ on molecules and -1.923na-5.465nβ on atomic species for G3(QCI/fu2)//B3PW91.The over-all accuracy is close to that of the G3 or G3/B3LYP method published in literature.The average absolute deviations on the G2-1 test set for these models are 4.37,4.39,4.06 and 4.02kJ mol-1,respectively,compared with 4.27 for G3 and 4.05 for G3//B3LYP on the same G2-1 test set.These results show further that the additivity or extrapolation approximation in the G3 theory is valid.However,the non-additive G3(QCI) procedures,which would be about four times of computer time consuming,in this work has been shown more reliable in the energy calculation of species at non-equiliberum geometry.Yhe maximum deveation of all of the four G3(QCI)'s occurs on the electronic affinity of the oxygen atom,e.g.-16.7 for G3(QCI/fu1) and -16.9kJ mol-1 for others compared with 17.6 for G3 and 18.4kJ mol-1 for G3//B3LYP on the electronic affinity of NH radical.In addition,this work suggests that the G3large basis sets for the first-row atoms are too small or the level of core-electronic energy calculations for the anions are too low since most of the larger errors are in the respective electronic affinities and this need to be further studied.