This paper studies the relationship between group size and informal risk sharing. It shows that under limited commitment with coalitional deviations, this relationship is theoretically ambiguous. It investigates this question empirically using data on sibship size of household heads and spouses from rural Malawi, exploiting a social norm among the main sample ethnic group to define the potential risk-sharing group. We uncover evidence of worse risk sharing of crop losses in larger potential risk-sharing groups, and rule out alternative explanations for the findings. A simple calibration exercise indicates that our empirical findings are consistent with the theory.