In this paper two different approaches to on-line optimization of batch processes are considered. In the first approach, a neural network strategy is utilized to compute, on-line, the optimal set-point trajectory to track. Once the optimal set-point trajectory is calculated a separate control algorithm can be used to track this trajectory. This approach takes into account batch-to-batch variations in initial conditions. In the second approach, tools from differential geometry are utilized to compute an optimal state feedback. This approach solves the optimization and control problem in one step and takes into account variations in possibly time-varying model parameters. These approaches are illustrated via several simulation examples.