We investigate the effects of group size and matching protocol in Tullock contest experiments. In a 2x2 factorial design we implement partner and random stranger matching protocols in contests between two and between three players. Group size does not affect overall absolute bid levels, but the rate of overbidding is lower in two-player groups across matching protocols. Matching protocol does not affect bidding behavior for three-player groups, but a partner matching reduces both the level and dispersion of bids in two-player groups. These results show the joint effect of group size and matching protocol, and suggest that identifiability in repeated play facilitates tacit collusion.