This paper studies the choice of network structures by criminal organizations in the presence of surveillance. A group of symmetric agents chooses a network structure and trades off better coordination, through centralized networks, with less exposure to the detection of its members, through decentralization. Bad surveillance technologies lead to smaller negative externalities between members which makes complete networks optimal. Improvements in surveillance lead to more decentralization, either through a split within the network or through more hierarchical structures. The type of decentralization depends on the value of indirect communication within the network.