High-speed milling (HSM) provides an efficient method for accurate discrete part fabrication. However, successful implementation requires the selection of appropriate operating parameters. Balancing the multiple process requirements, including high material removal rate, maximum part accuracy, chatter avoidance, and adequate surface finish, to arrive at an optimum solution is difficult without the aid of an optimization framework. Despite the attractive gain in productivity that HSM offers, full realization of the benefits is dependent on the proper selection of cutting parameters. Parameters selected must achieve the required productivity while maintaining an acceptable accuracy. Milling models are used to aid in the proper selection of these cutting parameters. They provide information on whether a cutting condition is stable and/or predict the surface accuracy. However, this selection is rather tedious, costly and time consuming and might not even provide an optimum solution. Parameters are selected based on experience until a point is found that provide the productivity and surface accuracy required. Difficulties encountered in this selection process include sensitivity of surface accuracy to cutting parameters, uncertainties in several parameters in the milling model and the computational effort needed to account for stability and surface accuracy. Therefore, balancing the multiple requirements, including high material removal rate, minimum surface location error and chatter avoidance, to arrive at an optimum solution is difficult without the aid of optimization techniques. In this dissertation a robust optimization algorithm that accounts for the inherent process uncertainty and surface location error sensitivity is developed. Two optimization criteria are considered, namely, surface location error and material removal rate under the stability constraint. The trade off curve of surface location error versus material removal rate is calculated for the mean values of input parameters, as well as for a confidence level in the stability boundary. An experimental validation of the robust optimization algorithm is also conducted, including an experimental validation of the variation of the cutting forces as a function of spindle speed. The confidence level in the axial depth limit and surface location error prediction is found using two methods: 1) sensitivity analysis; and 2) sampling methods. The sensitivity study highlights the most significant factors affecting process stability and surface location error. The effect of input parameters correlation is included in the confidence level predictions using Monte Carlo and Latin Hyper-Cube sampling methods.