There are significant incentives in controlling the end-use-properties in batch reactors to reduce the variability in the final product quality. The presence of model errors and other uncertainties poses a problem in achieving this objective using model-based control. A methodology to account for the model uncertainties in nonlinear model predictive control is presented here. This methodology also allows establishing a trade-off between robustness and optimality in the controller. This is based on determining the uncertainty in the predicted final values of the properties, in the form of elliptical confidence regions. The controller then ensures that the complete confidence region is within the target range for the product. A semiinfinite programming problem is solved to find the input values for the rest of the batch, utilizing techniques that reduce the complexity of the problem. A Parameter Adaptive Extended Kalman Filter is used to estimate the important unmeasured and observable properties. An emulsion polymerization process for styrene is chosen as a case study and the methods developed here are demonstrated on this example process.