Adult liking judgments for visual scenes are strongly influenced by the semantic content of images, more so than by lower-level visual features (e.g. the presence of specific colors or line types). This results in a strong degree of shared taste across different real-world scenes but highly idiosyncratic liking judgments for abstract images such as fractals and kaleidoscopic images that lack semantic content (Vessel & Rubin, 2010). This project seeks to understand how this adult-like pattern of shared taste emerges over the course of development. By measuring how long infants look at images, we infer their preferences, and use them to measure shared taste. Unlike adults, infants show a robust degree of shared taste for both real-world and abstract images. This provides evidence for a developmental shift in the level of representation that determines liking responses for sensory experiences: infant liking relies primarily on lower-level visual features while adult liking relies on higher-level semantic information and associations.