In this study, the predictive model of friction coefficient using cylindrical compression was constructed through combining the finite element method and neutral networks. Namely, the related data of the materials characters, cylinder compression bulging, and how they were associated with friction coefficient was obtained by the finite element method. Based on those analysis data, the relationship model, reflecting the relationship among the materials characters such as strength coefficient and strain-hardening exponent, the compression bulging such as reduction height, expanding in upper ending, expanding in bottom ending, maximum expanding in outside diameter and the friction coefficient in workpiece/die interface, was constructed. Finally, the cross verification between finite element analysis, prediction by neutral network model and the experiments of cylindrical compression testing and ring compression testing are repeatedly checked to ensure the accuracy and reliability of the constructed model. Results of the current study indicate that their errors are extremely limited, and the developed predictive system is reliable and feasible.