To export natural gas to overseas, it has to be liquefied and transported by the LNG tankers. Some LNG plants will receive their natural gas from different shale gas reservoirs so that one of the key challenges in developing a refrigerant mixture is the variation the of natural gas compositions. Previous attempts to create refrigerant mixtures for LNG plants were focused on implementing deterministic optimization methods. However, we demonstrate in this paper that these optimized refrigerants are so sensitive to natural gas composition that a slight variation in natural gas composition makes them unsuitable for the liquefaction process. To demonstrate the Gradient Assisted Robust Optimization technique capabilities, we developed a refrigerant mixture for propane precooled mixed refrigerant natural gas liquefaction cycles with an exaggerated variation in feed gas composition. This refrigerant is relatively unaffected by the variation of the natural gas compositions. We compared the performance of the new refrigerant to five other refrigerants found in literature and found the newly developed refrigerant is the only one satisfying the design constraints for all of the tested natural gas mixture compositions. This technique can be used for developing refrigerant for any LNG cycle that has a variation in feed composition. (C) 2016 Elsevier Ltd. All rights reserved.