Abstract: To examine when and why animals break up fights between others in their group, I modelled whether ‘winner’ and ‘loser’ effects might be one element driving the evolution of intervention behaviour. I considered one particular type of intervention: when the intervener simply breaks up fights between two others, but does not favour either party in so doing. When victories at time T + 1 are more likely given a victory at time T (i.e. winner effects), intervention is often favoured. Intervention is favoured in these circumstances because the intervening party in essence stops others from ‘getting on a roll’ and climbing up any hierarchy that exists. However, when loser effects alone are at work (defeats at time T + 1 are more likely given a defeat at time T), breaking up fights between others is never selected. If both winner and loser effects are operating simultaneously, then the likelihood of intervention behaviour evolving is a function of the relative strength of these two effects. The greater the winner effect relative to the loser effect, the more likely intervention behaviour is to evolve.