In this paper, we present a novel method to estimate color overprints from primary color spectra only. The method consists in building a mapping between primary color spectra and their combinations, based on a training set. The mapping is data-driven, without relying on assumptions based on classic theoretical models. By using this method, the number of prints and measurements for printer characterization is greatly reduced, which is of particular importance if there are many ink candidates to be combined. We tested the method on recently released Latex-based ink-jet ink, for different pigment and latex loads. The mean simulation accuracy error was below 2 dE, and gamut estimation error below 2% error. For pigment kinds, mean accuracy was about 5 dE, and gamut estimation error about 6%. In all cases, this new method outperforms other overprint estimation methods, such as Kubelka-Munk.