Abstract Adaptive behavioural responses often depend on qualities of the interacting partner of an individual. For example, when competing for resources, an individual might be better off escalating fights with individuals of lower quality, while restraining from fighting individuals of higher quality. Communication systems involving signals of quality allow individuals to reduce uncertainty regarding the fighting ability of their partners and make more adaptive behavioral decisions. However, dishonest individuals can destabilize such communications systems. An open question is whether cognitive mechanisms, such as learning, can maintain the honesty of signals, thus favoring their evolutionary stability. We present evolutionary simulations where individuals can produce a signal proportional to their quality and learn along their lifetime the best response to the signals emitted by their peers. Our simulations replicate previous results where the handicap principle mediates the evolution of signals. In our simulations learning on the receiver side can mediate the evolution of signals of quality on the sender side. When the cost of the signal is proportional to the quality of the sender, all individuals in populations are honest signalers. In contrast to traditional models which predict the absence of signals when the cost is not proportional to the quality of the signaler, our model revealed that learning facilitates the evolution of a polymorphism in which populations comprise both honest and dishonest signalers. We argue that learning may have a role in the evolution and dynamics of a wide range of communication systems and more generally in behavioral responses.