Sexual preferences in animals are often skewed toward mates with exaggerated traits. In many vertebrates, parents provide, through the learning process of "sexual imprinting," the model for the later sexual preference. How imprinting can result in sexual preferences for mates having exaggerated traits rather than resembling the parental appearance is not clear. We test the hypothesis that a by-product of the learning process, "peak shift", may induce skewed sexual preferences for exaggerated parental phenotypes. To this end, zebra finch (Taeniopygia guttata) males were raised by white parents, with beak color as the most prominent sexual dimorphism. We manipulated this feature with nail varnish. At adult age, each male was given a preference test in which he could choose among eight females with beak colors ranging from more extreme on the paternal to more extreme on the maternal side. The males preferred females with a beak of a more extreme color than that of their mothers, i.e., they showed a peak shift. Sexual imprinting can thus generate skewed sexual preferences for exaggerated maternal phenotypes, phenotypes that have not been present at the time of the learning. We suggest that such preferences can drive the evolution of sexual dimorphism and exaggerated sexual traits.