We consider in this work the interworking between multiple wireless access networks and study the multi-homed users' performance for different 4G/WiFi multi-homing techniques. We specifically compare two points of view for multihoming: network centric - wherein the scheduler is controlled by the network - versus user centric - where each user chooses a splitting policy for its packets. In the former, we study two proportional fairness strategies: a global one applied to the whole network, and a local one applied to each wireless access network apart. In the user centric approach, we study user policies based on the information received from the network and consider two variants: a simple one in which the user has only information on the peak rates of the radio interfaces and an optimized selfish policy in which the user receives complete information on the interfaces peak rates and traffic intensities. We then prove that this optimal selfish strategy achieves a global optimum for the system. Our numerical results show that network centric strategies are better than user centric ones in terms of achievable throughput but worst in terms of computational complexity