We present experimental evidence on favouritism practices. Children compete in teams in a tournament. After the first round of a real effort task, children indicate which group member they would prefer to do the task in the second round, for the benefit of the team. Friends are much more likely to be chosen than others after controlling for performance. We also find that children who are favoured by their friend subsequently increase performance. Consequently, favouritism does not hurt efficiency. These results show the importance of observing performance ex post in order to properly evaluate the efficiency implications of favouritism.