Quark/gluon discrimination is a key goal in jet substructure, with numerous potential applications in LHC searches for physics beyond the standard model. In order to make quark/gluon tagging a more robust tool, one wants first-principles QCD calculations to predict the degree of quark/gluon separation. To date, the best performing methods are based on Poisson-like counting observables which are not calculable using perturbation theory. Conversely, the most analytically tractable methods are based on Casimir-scaling observables which do not yield the best discrimination power. In this talk, I introduce a new quark/gluon discriminant that bridges this gap, yielding both excellent tagging performance and a high degree of analytic control.