A systematic approach to the analysis and design of a class of large dynamical systems is presented. The approach allows decentralized control laws to be designed independently using only local subsystem models. Design can be conducted using standard techniques, including loopshaping based on Nyquist and Popov plots, H-infinity methods, and mu-synthesis procedures. The approach is applied to a range of network models, including those for consensus, congestion control, electrical power systems, and distributed optimization algorithms subject to delays.