A method for designing the optimal working fluid mixture of organic Rankine cycles (ORC), which is robust against uncertainty in its composition, is developed. The objective function for designing the working fluid is formulated using statistical moments to achieve high performance even when the composition deviates from the nominal value. To obtain the statistical moments with a small number of simulations, nonintrusive polynomial chaos expansion (PCE) is employed. A Bayesian algorithm is adopted to solve the optimization of which objective function value is computed by PCE. The proposed method allows one to design robust working fluids, which cannot be performed in a reasonable time with Monte Carlo simulations, because of heavy computational efforts. It has been shown that changes in composition can have serious consequences to stable operation of ORC, and the working fluid designed by the proposed method allows flexible ORC operation, despite the existence of uncertainty in the composition.