Public good experiments are usually conducted in the laboratory using the voluntary contribution mechanism (VCM). We use a real public good (CO2 emission rights) with comparatively large groups (up to 60 subjects) and compare the results with corresponding experiments based on the VCM. Our treatment variations include group size and the feedback that subjects receive after each round. Our data reveal that, in contrast to the VCM, subjects dealing with a real public good relate their decisions to those of their group members in the lab only in the first rounds and do not change their contributions thereafter. These observations suggest that, when faced with a real public good, subjects use the whole group, including “players” outside the lab, as a main reference point. In this way, contributions made to the real public good in the lab tend to depend on the salience of mutual cooperation benefits existing outside the lab. Our findings help to understand the transferability of previous laboratory research on the VCM to real public good provision.