Group housing is often assumed to improve the welfare of gregarious species. Whether this is actually the case depends on the advantages (e.g. more opportunity for social and locomotor behaviour) and disadvantages (e.g. increased fighting and wounding) induced by the specific housing type. We evaluated the effects of a semi-group system (grouping four does for half of each reproductive cycle) on welfare by comparing it to single-doe cages. Compared to this control, our semi-group system provided more total space when does were grouped and more space per doe (a confounding deemed necessary to avoid overt aggression). Thus, the results should be interpreted as a systems comparison. In each of the four experimental cycles semi-group does were housed separately for 21 days around parturition and housed in newly assembled groups for the next 21 days. Behaviour was observed in semi-group and single-doe systems immediately after the second time semi-group does were mixed, and during five timeslots divided over the second experimental cycle. Skin lesion and weight loss were determined in each cycle. Adrenal weight was measured post-mortem. Semi-group systems with different floor types were included but floor type effects were scarce and semi-group systems were therefore treated as one category. In the timeslot subsequent to mixing semi-group does spent a greater percentage of their time on locomotion (4.3 vs. 0.7%, P < 0.01) and social sniffing/grooming (1.4 vs. 0%, P < 0.01) than does in single-doe cages. Such differences also occurred in later timeslots, but were much smaller (e.g. midnight locomotion D12: 0.8 vs. 0.2%, P < 0.05, midnight social sniffing/grooming D12: 0.4 vs. 0%, P < 0.01). Attacking/chasing followed a similar pattern (following mixing: semi-group 5.3% vs. single 0%, P < 0.01; midnight D12: 0.01 vs. 0%, P < 0.10). A high percentage of semi-group does were slightly (58%) or severely (20%) wounded. Semi-group does spent a smaller percentage of the timeslot following mixing in bodily contact with adults than does from single-doe housing (who could only make contact through the wire walls, 1.6 vs. 11.8%, P < 0.01). Even 12 days after mixing the percentage of time semi-groups spent in bodily contact did not exceed that in singles (P > 0.10). In experimental cycle one only, semi-group does lost more weight during late lactation than singles (192 vs. 10 g, P < 0.01). Adrenal weights did not differ between systems (P > 0.10). Further research will be needed to design semi-group systems with a more favourable balance between advantages and disadvantages.