For the nonlinear multi-agent systems with delay, an iterative learning control law was designed in order to achieve the stable formation of multi-agent systems.The dynamic characteristics of a single agent were established, and the problem of formation control of multi-agent was transformed into a tracking problem according to the topology structure of multi-agent.A unified iterative learning controller was designed for each agent.According to the different dynamic characteristics of each agent, a suitable learning gain matrix was selected.Then the convergence of the proposed iterative learning law was proved based on the λ norm theory.Due to the relaxation of initial value in the iterative learning control algorithm, the stability of the control can be achieved in any constant initial condition.The algorithm proposed can meet the formation control requirement for each agent having random initial position.Finally, the simulation results were given to illustrate the effectiveness of the proposed method.