An effective unsteady aerodynamic-optimization method of airfoil for cycloidal propellers is developed. Firstly, the class-function/shape-function transformation parameterization method is adopted to describe the airfoil profile with different geometrical parameters. Secondly, the sample points in the design space are selected using the optimal Latin hypercube design methodology. Thirdly, the actual aerodynamic performance of cycloidal propellers is simulated by using the time-dependent incompressible Reynolds-averaged Navier–Stokes computational fluid dynamics method and the sliding-mesh technique, which are verified by validation cases on an oscillating NACA 0012 airfoil and the baseline cycloidal propeller. Then, to replace the time-consuming computational fluid dynamics solver, kriging surrogate models are established using the sample data. Finally, the genetic algorithm is adopted to determine the optimized airfoil in the design space based on the constructed kriging models. Optimization is conducted on a micro- four-bladed cycloidal propeller with the NACA 0015 airfoil, and the results show that the figure of merit of the cycloidal propeller is increased by 18.31% at the design rotation speed. Although the optimization is conducted only at the design rotation speed, an appreciable improvement of the figure of merit can be achieved for a wide range of rotation speeds.