There is a large literature on the types of dialogue features that are theorised to be productive for learning; however, there is very little evidence for whether these forms of dialogue are more common, applicable or impactful in certain subjects. This mixed-methods study explores how dialogue unfolds in different curriculum contexts. Classroom dialogue was examined through naturalistic observations of English, mathematics and science lessons in primary schools in England. The Cambridge Dialogue Analysis Scheme (Vrikki et al., 2019) was used to gather the frequency of key dialogue features in 72 lessons (ages 10-11). Statistical analysis identified that reasoned dialogue is more frequent in mathematics and both English and mathematics have a proclivity towards more elaborated dialogue in comparison to science. Furthermore, attainment and subject dialogues were examined in 63 lessons through multi-level modelling, revealing that elaboration (in conjunction with reference to the wider context) in English was associated with student attainment in reading and spelling, punctuation and grammar. The development of a typology of activity types relevant to each subject under consideration added an additional layer of coding that enabled the exploration of each activity type for its dialogicality, in an attempt to explore possible partial explanations of subject differences in dialogue. This analysis of 63 lessons revealed stark differences in the durations of dialogic activity types in each subject. Finally, latent thematic analysis of 12 teachers’ interviews have shed light on why the nature of dialogue may differ between the core disciplines at primary level; the centrality of reasoning in the mathematics curriculum and the marginalisation of primary science appear prominently in teachers’ accounts. The latter, in particular, has important implications for education policy; revaluation of the division of teaching time between the core subjects and the breadth and depth of topics covered in the primary science curriculum is recommended.