Abstract: Group housing of horses is not widely applied in practice despite the welfare advantages of keeping animals socially rather than individually. In particular, concerns have been raised about the possible increased risk of injury and how to introduce a new horse into an established group. This study investigated two hypotheses: (1) pre-exposure of young horses in neighbouring boxes reduces the frequency of aggressive interactions when the same horses are subsequently put together in a paddock compared to horses without this previous box experience, (2) the occurrence of aggressive behaviour, in particular contact aggression in the paddock can be predicted after observing the horses' social interactions in neighbouring boxes. Danish Warmblood mares (n = 20), 2 years old, were kept in two groups of 10 horses. In total, 60 encounters were arranged whereby each horse was confronted pair-wise with six horses from the other group, three according to each treatment: treatment I--box (B) and subsequent paddock meeting (BP), and treatment II--only paddock meeting (P). Horses met in neighbouring boxes for 5 min and together in the same paddock for 10 min. The frequencies of aggressive and non-aggressive interactions were analysed from video recordings. Total aggression levels between BP and P did not differ, but [`]contact aggression', i.e. bite, kick, strike, push, tended to be lower in BP compared to P (median BP = 1, P = 2; p = 0.083) and there were less bites in BP than P (median BP = 0, P = 1; p = 0.050). Frequencies of [`]non-aggressive' interactions, e.g. friendly approach, nasal sniff, were lower in BP than P (median BP = 2.5, P = 10; p < 0.01). Results further revealed that [`]bite threat' performed in boxes correlated with [`]contact aggression' in the paddock (r = 0.46, p = 0.011). In conclusion, pre-exposure of young horses in neighbouring boxes may reduce [`]contact aggression', especially biting, in the paddock and [`]bite threat' shown in boxes may help to predict contact aggression when horses are later turned out together. The reduced non-aggressive interactions in the paddock in the BP test were probably a consequence of horses having exchanged these behaviours in the preceding B test. Exposing young horses in boxes next to each other may be a helpful tool before mixing them because horses meet in a safe environment that could assist in reducing the type of aggression where horses are most at risk of being injured.