Abstract: For the downlink of muti-user cellular networks with two users per cell, an interference alignment algorithm based on interference graph is proposed. We first select five adjacent cells to be partitioned into a small cluster from the interference graph which represents the distribution of inter-cell interference. For the symmetric?and asymmetric?channels, we design the receive beamforming matrices to align all the inter-cell interference or group inter-cell interference into a subspace separately. Then, consider the inter-clusters interference, we can also align it to inter-cell interference subspace. And the transmit beamforming matrices orthogonal to the interference subspace which is figured out in clusters. Finally, we can achieve the interference alignment of all the active cells in the cellular networks by clusters’ translation. The proposed algorithm can reduce both the transmitting and receiving?antennas and the computational?complexity. Simulation results show that the proposed algorithm eliminates the inter-cluster interference, inter-users interference and inter-clusters interference simultaneously. It effectively improves the capacity of the cellular networks.