Reporting private information is a key part of economic decision making. A recent literature has found that many people have a preference for honest reporting, contrary to usual economic assumptions. In this paper, we investigate whether preferences for honesty are malleable and what determines them. We experimentally measure preferences for honesty in a sample of children. As our main result, we provide causal evidence on the effect of the social environment by randomly enrolling children in a year-long mentoring programme. We find that, about four years after the end of the programme, mentored children are significantly more honest.