Accurate simulation of intracellular biochemical networks is essential to furthering our understanding of biological system behavior. The number of protein complexes and of chemical interactions among them has traditionally posed significant problems for simulation algorithms. Here we describe an approach to the exact stochastic simulation of biochemical networks that emphasizes the contribution of protein complexes to these systems. This simulation approach starts from a description of monomeric proteins and specifications for binding, unbinding and other reactions. This manageable specification is reasonably intuitive for biologists. Rather than requiring the inclusion of all possible complexes and reactions from the outset, our approach incorporates new complexes and reactions only when needed as the simulation proceeds. As a result, the simulation generates much smaller reaction networks, which can be exported to other simulators for further analysis. We apply this approach to the automatic generation of reaction systems for the study of signal transduction networks.