There has been a great deal of interest in whether animals use trait symmetry as a visual cue to mediate behavioural interactions. In bilaterally symmetric traits, small asymmetries (termed fluctuating asymmetry) appear due to increased developmental stress and/or genes for poor developmental homeostasis. Hence, researchers have hypothesized that symmetry can reveal the developmental history and, perhaps, fitness of an individual and this is why symmetry preferences have been observed. However, an additional theory suggests that symmetry could be preferred merely because it represents the average expression of bilateral traits. Animals can learn to respond to signals by generalizing (or averaging) stimulus sets. As the average expression of a trait showing fluctuating asymmetry is zero asymmetry, theory predicts that animals could develop a symmetry preference as a by-product of learning. Here, we test this prediction empirically with European starlings (Sturnus vulgaris) and show that symmetry preferences can emerge as an outcome of generalized learning processes. Our results indicate that symmetry does not initially need to be associated with fitness to be an apparent cue in behavioural interactions and that symmetry preferences observed in nature could be independent of any putative fitness relationships.