This contribution summarizes a methodology for simulation optimization assuming some simulation inputs are uncertain. This methodology integrates Taguchi’s worldview (distinguishing between decision and environmental inputs), metamodeling (either Response Surface Methodology or Kriging), and mathematical programming. Instead of Taguchi’s statistical designs, this contribution uses Latin Hypercube Sampling for the environmental inputs. Mathematical programming is used to estimate the decision inputs that minimize the mean output, subject to a threshold for the standard deviation of the simulation output. Changing that threshold gives the estimated Pareto frontier. Confidence regions for the Pareto-optimal solution based on that frontier can be estimated through bootstrapping. This methodology is illustrated through Economic Order Quantity (EOQ) simulations.