Assortative mating displays and/or preferences can be affected by learning across a wide range of animal taxa, but the specifics of how this learning affects speciation with gene flow are not well understood. We use population genetic models with trait learning to investigate how the identity of the tutor affects the divergence of a self-referent phenotype-matching trait. We find that oblique learning (learning from unrelated individual of the previous generation) and maternal learning mask sexual selection and therefore do not allow the maintenance of divergence. In contrast, by enhancing positive frequency-dependent sexual selection, paternal learning can maintain more divergence than genetic inheritance, but leads to the loss of polymorphism more easily. Furthermore, paternal learning inhibits the invasion of a novel self-referent phenotype-matching trait, especially in a large population.