The main objective of this study was to investigate the effects of a programme providing cross-age peer-mediation on mediation teaching style of mediators and learners in a learning situation. A second objective was to investigate the effects of the mediators' and learners' cognitive level on different criteria of mediation. The sample was composed of 89 third graders (mediators) and 89 first graders (learners) coming from three third grade classes and three first grade classes in a primary school. The third graders were assigned to experimental (n = 43) and control (n = 46) groups. Each pair of mediator and learner was matched in a counterbalanced design (2 × 2) according to their cognitive level (high versus low), based on their score on the Raven's Matrices. The experimental third graders received the Peer-Mediation for Young Children (PMYC), which is a training programme based on the theoretical approaches of Vygotsky and Feuerstein. The control children received a general preparation for peer assisted learning. Following the PMYC, all children participated in a peer-mediation condition, which was videotaped for 25 minutes and subsequently analysed by the Observation of Mediation Instrument. The findings showed that the experimental children received significantly higher mediation scores than the control children. Mediators' cognitive level was not related to the level of mediation. However, the learners' cognitive level was found to effect mediation for meaning of the mediator: request for meaning from learners and giving of meaning from mediator were higher in interactions of mediators with learners of low cognitive level. The findings are discussed with reference to Feuerstein's and Vygotsky's theories.