This paper presents a sequential set of tests for model-based Real-Time Optimization (RTO) system design, using the framework of the Design Cost Criteria to evaluate design alternatives. Design Cost is defined as the "lost profit" due to offset from the optimal plant operations and variance of the RTO system predictions of optimal operations. Necessary conditions for elimination of offset are presented, which can be implemented in a sequential fashion to concentrate increasing effort on only those models which have satisfied the previous conditions. A method for approximating the variance of the predicted plant optimum is developed. From these tools, a key parameter selection procedure is synthesized and illustrated with a reactor operations optimization example.