Social networks are an important component in understanding the decision to consume addictive substances. They capture the role of limited access, peer influence, and social acceptance and tolerance. However, despite the empirical evidence of their role, they have been absent from theoretical models. This paper proposes a mechanism through which agents can influence each other in their decision to consume an addictive good. An agent's decision is sensitive to her state of addiction as well as to the composition of her neighbourhood. The model is consistent with the empirical evidence that peer influence can work in both ways: influencing an individual to use and helping them to quit. The structure of the network has important implications on the outcome of agents' decisions as well as the effectiveness of policies aimed at limiting use of addictive substances through deterrence. I provide a network-based explanation of why usage rates can vary across otherwise similar agents and show how in some situations encouraging network ties can lead to lower use while in others it can have the opposite effect. Furthermore, I explore the effect of networks on diffusion of addiction and, using simulations, find that addiction spreads faster in an environment where there are few strong links than in one with many weak links.