Theory suggests that genetic polymorphisms in female mating preferences may cause disruptive selection on male traits, facilitating phenotypic differentiation despite gene flow, as in reinforcement or other models of speciation with gene flow. Very little experimental data have been published to test the assumptions regarding the genetics of mate choice that such theory relies on. We generated a population segregating for female mating preferences and male colour dissociated from other species differences by breeding hybrids between species of the cichlid fish genus Pundamilia. We measured male mating success as a function of male colour. First, we demonstrate that non-hybrid females of both species use male nuptial coloration for choosing mates, but with inversed preferences. Second, we show that variation in female mating preferences in an F2 hybrid population generates a quadratic fitness function for male coloration suggestive of disruptive selection: intermediate males obtained fewer matings than males at either extreme of the colour range. If the genetics of female mate choice in Pundamilia are representative for those in other species of Lake Victoria cichlid fish, it may help explain the origin and maintenance of phenotypic diversity despite some gene flow.