In this study, the finite element method was used to analyze comprehensively the effects of punch shape on forming the forging recess. Then, the polynomial network and genetic algorithm were combined to construct the predicted and designed system. Through this approach, we can predict the forging results formed from arbitrary shaped punches, and design the optimal punch shape for forging recess. Through the interactive verifies of modeling repeatedly, the errors resulted from modeling analysis, network prediction and genetic algorithm optimal design are extremely limited. Consequently, the predicted and designed approach of optimal punched shape for forming recess in this study could be extended to the design of more complicated and difficult formed forging die.