Identifying interaction partners, a particularly challenging task in the absence of an explicit communication system (e.g. vocalization), is key to understanding how information is distributed and processed within animal groups. Moreover, the interaction network regulates complex collective behaviours such as collective motion and predator detection. Despite the relevance of the interaction network, only a few experiments have been designed to decipher it. Moreover, the mechanism by which an individual selects its interaction neighbours remains largely unknown. Here, we investigated the interaction neighbour selection process, by studying the behavioural response of naive group members confronted with the controlled departure of a trained individual in herds of merino sheep, Ovis aries. By identifying the first individual that followed the trained individual, we inferred the probability of a group member choosing it as its interaction neighbour, assuming that this probability is modulated by either the relative distance, the relative distance rank or a combination of both. The underlying interaction network was constructed from this probability. The results provide unique, reliable information on the functional form of this probability, and thus on the interaction network and on how individuals interact within groups. The suggested method of working out the selection process can be easily implemented in other animal systems and contexts.