||Abstract There is increasing interest in keeping horses in groups, but progress is hampered by a lack of knowledge about which horses can and should be kept together. Therefore, our objective was to investigate the effect of group composition on the occurrence of injuries among horses, the ease of removing horses from groups and horses’ reactivity to a fearful stimulus. Using a matched case control design, 61 groups of horses were studied in Denmark, Norway, Finland and Sweden. They were allocated into groups of similar or different age and sex or where membership changed regularly or remained stable. Injuries were recorded before mixing the horses into treatment groups, the day after mixing and four weeks later. Reactivity of horses to a moving novel object and the behaviour of a horse being removed from its group and the reactions of other group members towards this horse and the handler were evaluated. It was hypothesized that a more socially variable group composition has beneficial effects on behaviour, ease of handling and reducing reactivity whereas frequent changes in group composition has negative consequences, resulting in more injuries. We found that differences in treatment effects were mainly related to breed, rather than group composition. Icelandic horses reacted less to the movement of the novel object (P = 0.007) and approached it more afterwards (P = 0.04). They also had fewer new injuries than warmbloods following mixing (P < 0.001) and fewer than all other groups 4 weeks later (P < 0.01). Most new injuries after mixing were minor and recorded on the horse’s head, chest, hind legs and rump. In conclusion, variations in sex and age composition of the group had little effect on injury level, reactivity and ease of handling compared to the general effect of breed. Concerns about the risk of severe injuries associated with keeping horses in groups are probably overestimated. Thus, we propose that horses can be successfully kept in groups of different sex and age composition.