Economists are increasingly interested in how group membership affects individual behavior. The standard method assigns individuals to "minimal" groups, i.e. arbitrary labels, in a lab. But real group often involve social interactions leading to social ties between group members. Our experiments compare randomly assigned minimal groups to randomly assigned groups involving real social interactions. While adding social ties leads to qualitatively similar, although stronger, in-group favoritism in cooperation, altruistic norm enforcement patterns are qualitatively different between treatments. Our findings contribute to the micro-foundation of theories of group preferences, and caution against generalizations from "minimal" groups to groups with social context.