Agent-based simulation for interaction behavior between group and work was explored from an evolutionary game-theoretical perspective.This paper develops a pay off-shared andunishment-shared game mode,l design evolution learning rules considering of historical information and decisioncharacteristicofneigh-bor,sand uses multiagent approach to represent groupwork.Based on class libof Repas,tweuseJava2 to program the multi-agent simulation system.Simulation results indicate that(1)the size of work group has a minor effect on cooperation trend and major effect on cooperation frequency of group, (2)total work pay off bhasa positive effect on the stability of work state of group.Group behavior is inastate of Tit-for-Tat when work costci sequivalent to work punishment, more player swant to cooperate when c>d, and vice versa, the more players wants to defec twhen c