Across a variety of domains there are experts who can discriminate between complex visual categories. Exemplar theory suggests that through exposure to multiple examples, people can learn the visual style of a category and use this experience to categorise new instances (Brooks, 1978). Our aim was to investigate whether novices learn to discriminate between artistic styles more effectively when presented with exemplars from different categories compared to exemplars from the same category. Drawing on exemplar theory and research on contrastive learning, we developed a novel training methodology. In our experiment (N = 90) we manipulated the number and type of exemplars presented during learning. There were two types of learning trials: paintings by the same artist (match) and paintings by different artists (nonmatch). Learning reflected the ability to successfully classify novel pairs of paintings as matches or non-matches. Transfer items included novel instances of new artists or novel instances of artists encountered during training. Transfer of learning was assessed on an immediate test, and retention was assessed one week later. We predicted that presenting multiple exemplars would improve learning, and that exemplars that highlight the variation between categories would be the most effective for learning (particularly for transfer to new artists) and retention. There was no evidence to suggest that presenting multiple exemplars improved learning of style, or that the type of exemplar influenced learning and there was no difference in transfer items. There was some evidence to suggest that participants were more accurate at discriminating the style of non-matches compared to matches, although this was partially driven by a slight conservative response bias. This research provides a starting point in developing an empirical basis for understanding how novices learn the style of complex categories through exposure to the variation within and between categories. Future research could replicate the current methodology using stimuli with reduced visual information, or by additionally providing classification rules.