Recent research demonstrates that both real-time variability in perceptual input and task demands influence young children’s word learning and categorisation. The current study extends these findings by testing both children and a dynamic field theory (DFT) computational model in a category labelling task. Specifically, children and the model were introduced to multiple category members that were either moderately or highly variable. Both children and the model were better able to learn category labels when the individual category members were moderately variable. Overall, these findings have implications for both our understanding of children’s categorisation and the use of computational models to investigate cognition more generally.