We study children’s cheating by conducting a field experiment in a local primary school. Children graded either their own or another student’s test, and they could cheat by misreporting the overall score. Unbeknownst to them, the test-taker’s original answers were recorded by carbonless copy paper. As expected, we find that children were generally more likely to cheat for themselves compared to cheating for others. To investigate cheating for others, we vary whether children graded their friend or an acquaintance and whether the grading pairs could discuss the test while grading. For the friend, children cheated little with or without discussion. For the acquaintance, they also rarely cheated without discussion; but with discussion, they cheated frequently, nearly as much as when grading themselves. We discuss implications of these findings on social cheating for theories about reciprocity and reputation.