A new global fuzzy iterative learning scheme is proposed for nonlinear multi-agent systems with unknown dynamics. Unlike the traditional design scheme where the fuzzy systems are used as the feedback compensators, the fuzzy systems are used as the feedforward compensators to describe the unknown dynamics, which avoids the restriction on the states of the control systems. In this scheme, we design a hybrid fuzzy adaptive learning controller according to the characteristics of the network structure. On this basis, using the Nussbaum function, this paper extends the above global fuzzy iterative learning scheme to solve the consensus control problem of multi-agent systems with unknown control directions over the iterations. Finally, the effectiveness of the above hybrid learning protocols is verified through simulations.