Within system dynamics, optimization has played an important role in identifying the best range of parameter values for policies in any given model. Optimal solutions focus on discovering the best combination of model parameters, within a fixed policy equation structure, that maximize or minimize a payoff function. This paper presents a new optimization approach for system dynamics. It enables decision makers to vary policy equation structures during the optimization process. The resulting optimization approach-based on genetic algorithms-can explore the search space in order to discover the best combination of parameters and equation-based strategies for a given system dynamics problem. The approach is best suited to the class of system dynamics problems that are agent-based, and the work is evaluated using a case study based on the four-agent beer game. Copyright (c) 2008 John Wiley & Sons, Ltd.