The canonical transformation–cluster expansion formalism is used to generate the effective valence shell Hamiltonian for carbon. Hydrogenlike orbitals defined by an effective nuclear charge parameter 𝒵 are used to span the core (K shell), valence (L shell), and excited (3⩽n⩽9) spaces. The effective Hamiltonian containing one- and two-body interactions is diagonalized on the Nv-particle valence space to yield the low-lying excitation spectrum. Considering alternative approximations to carry out the calculations, we indicate the importance of including the two-body pair potential as well as the single particle operators in the generator of the canonical transformation. Upon doing this, good agreement with experiment is obtained for the lowest valence shell transitions over a wide range of 𝒵. For certain physically reasonable 𝒵 the entire valence shell experimental excitation spectrum can be accurately reproduced. In contrast, a ’’zeroth order’’ effective Hamiltonian using only the ’’charge cloud’’ of the core to modify the one-body potential and using e2/r12 for the two-body potential fails to reproduce the spectrum for any choice of 𝒵. We compare our effective Hamiltonian with those currently used in semiempirical calculations. We find the splitting in the 2s–2p orbital energies as well as the ’’strength’’ of the two-body interaction to be substantially reduced from those corresponding quantities occurring in the zeroth order effective Hamiltonian. These results generally fulfill the semiempirical expectations. We thus consider the procedure as a method for obtaining the valence shell Hamiltonian for molecules. Preliminary results for ethylene are presented and discussed.