This paper describes a neural network based modeling scheme of parameter function modeling that captures the inherently nonlinear dynamics of fed-batch bioreactor systems. A neural network model in conjunction with basic material and energy balances provides a better representation of the system dynamics than a purely empirical model. The combined balance relation and neural network parameter function model can be used with dynamic programming optimization to generate optimal operational policies. The results of applying this method to two fed-batch bioreactor systems are presented and shown to agree well with exact optimal strategies.