We analyze the friendship paradox via a laboratory experiment. The paradox comes from the fact that people’s friends have more friends than people do, on average. Thus, if people fail to account for the fact that their friends are not typical representatives of the population, then they can overestimate the society’s norm of behavior, potentially leading to an upward bias in behavior. To test the paradox in the lab, we put subjects in a network in which there are popular agents and poorly connected ones, and we vary whose actions subjects observe. The paradox should only appear when subjects’ payoffs depend on the actions of the whole society and they only see the actions of their neighbors. Our preliminary results find a significant upward bias relative to the equilibrium in the situation in which players do not know the network, but a more limited bias when they do know the entire network. We also find that the bias is larger for the behavior of the more highly-connected subjects.