For uplink multi-user massive MIMO systems with hundreds of antennas at the base station (BS), minimum mean square error (MMSE) signal detection method is near-optimal due to the asymptotically orthogonal channel, but it involves matrix inversion with high complexity. In this paper, a lowcomplexity signal detection method in an iterative way is proposed to avoid the complicated matrix inversion. The convergence proof and complexity analysis of the proposed method are provided in detail. The analysis shows that the proposed method can effectively reduce the complexity by one order of magnitude. Simulation results demonstrate that the proposed method outperforms recently proposed Neumann series approximation method. Near-optimal performance of the MMSE method can be achieved with only a small number of iterations.