We develop a key-player model by allowing for link formation so that when a person is removed from a network the other individuals can form new links while still optimally providing crime effort. We then put our model to the test, using data on adolescent delinquents in the United States, and provide new results regarding the identification of peer effects. This is done by a structural estimation and simulation of our model. Compared to a policy that removes
randomly delinquents from the network, a key player policy engenders a crime reduction that can be as large as 35 percent. We discuss how to implement the key-player policy in the real world, primarily within criminal networks, but also within financial, R&D, development, political and tax-evasion networks.