Background. This article describes the outcomes of a longitudinal, multilevel observational study in which the relationship between the processes of co-operation and giving explanations was compared between classes trained in communication skills and classes that were untrained. Aims. This study examined the effects of training in basic communication skills on the processes of co-operation and giving explanation in co-operative groups. In particular, the study investigated: (a) How the processes of cooperation and giving explanations develop over time: (b) How the underlying processes of co-operation and giving explanations are related: (c) Which factors at the student and class level facilitate or hinder these processes. Sample. The study involved 192. Year 6 primary school children (mean age = 134.3 months) who worked in four-person, mixed ability, gender-balanced groups on a social studies unit of work for 12 weeks. Results. The students in the trained classes were more co-operative and gave more explanations to each other than their peers in the untrained classes. Giving solicited explanations could be traced back to co-operation and this process was enhanced by ability. In other words, the higher the individual ability levels of the students, the more co-operation was transformed into giving solicited explanations. Furthermore, over and above the effects of student ability, the higher the class' ability level, the more explanations the students gave. Conclusion. The results of this study are useful in explaining why high-ability students benefit more from co-operative learning than low-ability students and why solicited explanations are more effective than unsolicited explanations. By opening the black box of co-operative learning, our analysis enables us to attribute the effects of co-operative learning to peer interactions with more able peers and this is promoted by class ability level. These factors have been mentioned in the literature as a possible explanation of 'contextual effects' but not investigated empirically.