In order to design a horn for small hole extrusion machining with ultrasonic vibration,an optimization method of composite horn was proposed. Three surrogate models,Polynomial Response Surface,Kriging and Support Vector Regression,were employed to predict the performance parameters of the horn according to its dimension parameters. The accuracy of the surrogate models were compared. Based on genetic algorithm and the model with the higher accuracy,the optimization of the composite horn was investigated. The simulation results confirm the validity of the optimization method. The experimental results show that the horn meets the requirements of use and the machining effect is ideal.