Vocal learning has evolved in several groups of animals, yet the reasons for its origins and maintenance are controversial, with none of the theories put forward appearing to apply over a broad range of species. The theory of gene–culture coevolution is applied to this problem taking the specific case of the maintenance of song learning in birds. The interaction between genes underlying the filter for recognizing and learning conspecific song and the culturally transmitted songs themselves sets up an evolutionary force that may maintain vocal learning. We evaluate this hypothesis using a spatial simulation model. Our results suggest that selection that would maintain song learning exists over a wide range of conditions. Song learning may persist due to an evolutionary trap even though the average fitness in a population of learners may be lower than in a population of non–learners.