There is conflicting evidence on whether collaborative group work leads to improved classroom relations, and if so how. A before and after design was used to measure the impact on work and play relations of a collaborative learning programme involving 575 students 9-12 years old in single- and mixed-age classes across urban and rural schools. Data were also collected on student interactions and teacher ratings of their group-work skills. Analysis of variance revealed significant gains for both types of relation. Multilevel modelling indicated that better work relations were the product of improving group skills, which offset tensions produced by transactive dialogue, and this effect fed through in turn to play relations. Although before intervention rural children were familiar with each other neither this nor age mix affected outcomes. The results suggest the social benefits of collaborative learning are a separate outcome of group work, rather than being either a pre-condition for, or a direct consequence of successful activity, but that initial training in group skills may serve to enhance these benefits. (Contains 6 tables.)