Controlling employees can have severe consequences in situations that are not fully contractible. However, the perception of control may be contingent on the nature of the relationship between principal and agent. We, therefore, propose a principal-agent model of control that takes into account social identity (in the sense of Akerlof and Kranton, 2000, 2005). From the model and previous literature, we conclude that a shared social identity between the principal and agent has both a cognitive, that is, belief-related, and a behavioral, that is, performance-related, dimension. We test these theoretical conjectures in a labor market experiment with perfect monitoring. Our ndings confirm that social identity has important implications for the agent's decision-making. First, agents who are socially close to the principal (in-group) perform, on average, more on behalf of the principal than socially distant (no-group) agents. Second, social identity shapes the agent's subjective expectations of the acceptable level of control. In-group agents expect to experience less control than no-group agents. Third, an agent's reaction to the monitoring level she eventually faces also depends on social identity. If the experienced level of control is lower than the expected control level, that is, the agent faces a positive sensation, the increase in performance is less pronounced for in-group agents than for no-group agents. In the case of a negative sensation, however, in-group agents react stronger than no-group agents. Put differently, being socially distant from the principal amplies the performance-enhancing effect of a positive control surprise and mitigates the detrimental performance effect of a negative surprise.