Iconicity is the property whereby signs (vocal or manual) resemble their referents. Iconic signs are easy to relate to the world, facilitating learning and processing. In this study, we examined whether the benefits of iconicity would lead to its emergence and to maintenance in language. We focused on shape iconicity (the association between rounded objects and round-sounding words like "bouba" and between spiky objects and spiky-sounding words like "kiki") and motion iconicity (the association between longer words and longer events). In Experiment 1, participants generated novel labels for round versus spiky shapes and long versus short movements (Experiment 1a: text, Experiment 1b: speech). Labels for each kind of stimulus differed in a way that was consistent with previous studies of iconicity. This suggests that iconicity emerges even on a completely unconstrainted task. In Experiment 2 (Experiment 2a: text, Experiment 2b: speech), we simulated language change in the laboratory (as iterated learning) and found that both forms of iconicity were introduced and maintained through generations of language users. Thus, we demonstrate the emergence of iconicity in spoken languages, and we argue that these results reflect a pressure for language systems to be referential, which favors iconic forms in the cultural evolution of language (at least up to a point where it is balanced by other pressures, e.g., discriminability). This can explain why we have iconicity across natural languages and may have implications for debates on language origins. (PsycInfo Database Record (c) 2022 APA, all rights reserved).