The structured system for peer assisted learning in writing named Paired Writing (Topping, 1995) incorporates both metacognitive prompting and scaffolding for the interactive process. This study sought to evaluate the relative contribution of these two components to student gain in quality of writing and attitudes to writing, while controlling for amount of writing practice and teacher effects. Participants were 28 ten‐ and eleven‐year‐old students forming a problematic mixed ability class. All received training in Paired Writing and its inherent metacognitive prompting. Students matched by gender and pre‐test writing scores were assigned randomly to Interaction or No Interaction conditions. In the Interaction condition, the more able writers became ‘ tutors’ for the less able. In the No Interaction condition, the more able writers acted as controls for the tutors and the less able as controls for the tutees. Over six weeks, the paired writers produced five pieces of personal writing collaboratively, while children in the No Interaction condition did so alone. On pre‐ and post‐project analyses of the quality of individual writing, all groups showed statistically significant improvements in writing. However, the pre‐post gains of the children who wrote interactively were significantly greater than those of the lone writers. There was some evidence that the paired writers also had more positive self‐esteem as writers. The operation and durability of the Paired Writing system are discussed.