This research adopted observational and experimental paradigms to investigate the relationships between components of emotion knowledge in three‐ to four‐year‐old children. In Study 1, 88 children were assessed on the Emotion Matching Task (Morgan, Izard, & King), and two tasks requiring the generation of emotion labels and causes. Most tasks were significantly associated with age and language ability, and similar tasks were significantly but moderately correlated. In Study 2, 58 of these children were allocated to one of three conditions: emotion cause talk, in which they received four sessions of training focusing on emotion labels and causes; non‐emotion cause talk, focusing on causal relationships in general; or were allocated to a no‐training control. Children in the emotion causes condition showed improvement in their use of emotion labels but not other components of emotion knowledge. The findings suggest that, in three‐ to four‐year‐old children, emotion knowledge is constituted by related but separable components and that training in emotion language targets separate rather than all aspects of emotion knowledge.