Starting from a comparative study of various methods for uncertainty propagation, this paper presents a novel reduced quadrature technique to be used in gradient-based robust design optimization of aerodynamic shapes. The accuracy and computational efficiency of the method are investigated by means of mathematical analyses and numerical examples. The method is then applied to the robust design of airfoils under probabilistic uncertainty. It is shown that the solutions obtained through the proposed method can outperform those obtained through linearization, without any significant increase in computational cost.