We describe a model for spectral tuning in red fluorescent proteins (RFPs) based on the relation between an electronic structure descriptor, the dipole moment variation upon excitation (DMV), and the excitation energy of a protein. This approach aims to overcome a problem of accurate prediction of excitation energies in RFPs, which span a very narrow window of band maxima. The latter roughly corresponds to an energy gap of 0.1 eV, which is comparable with typical errors in calculations of the excitation energy by conventional quantum chemistry methods. In this work, we demonstrate a strong quantitative correlation between DMV values, obtained computationally with modest efforts, and excitation energies ΔEex at the experimental excitation band maxima for a series of RFPs with the bands between 570 and 605 nm. Protein models are constructed by motifs of the relevant crystal structures, and atomic coordinates are optimized in quantum mechanics/molecular mechanics (QM/MM) calculations with QM-subsystems composed of large chromophore-containing regions. DMV values are evaluated with the electron density computed at the TDDFT level using several functionals and basis sets. We demonstrate that the results obtained with the CAM-B3LYP, BHHLYP and M06-2X functionals demonstrate favorable correlations between DMV and ΔEex with the mean absolute error less than 0.01 eV. Taking into account the solid theoretical grounds of the relation between the DMV and the excitation energy in fluorescent proteins, the described modeling strategy presents a rational tool for spectral tuning in these efficient markers for in vivo imaging.