A glass composition variation study (CVS) for high-level waste (HLW) stored in Idaho is being statistically designed and performed in phases over several years. The purpose of the CVS is to investigate and model how HLW-glass properties depend on glass composition. The resulting glass property–composition models will be used to develop desirable glass formulations and for other purposes. Phases 1 and 2 of the CVS have been completed and are briefly described. This paper focuses on the CVS Phase 3 experimental design, which was chosen to augment the Phase 1 and 2 data with additional data points, as well as to account for additional glass components not studied in Phases 1 and/or 2. In total, 16 glass components were varied in the Phase 3 experimental design. The paper describes how these Phase 3 experimental design augmentation challenges were addressed using the previous data, preliminary property–composition models, and statistical mixture experiment and optimal experimental design methods and software. The resulting Phase 3 experimental design of 30 simulated HLW glasses is presented and discussed.