In this paper, we propose a sparsity-aided iterative receiver for large scale under-determined multiple-input multiple-output (UD-MIMO) systems. The proposed scheme is motivated by the fact that most conventional receivers produce a sparse residual error vector, which is the difference between the actual transmitted symbol vector and the estimated one. The sparse feature of the residual error vector is utilized to locate the support set of the erroneously detected symbols by using the compressive sensing (CS) framework. We can then remove the effects of the correctly detected symbols and only update the soft information of the symbols with detection errors. Since the number of error symbols is usually much less than that of receive antennas, the sparsity-aided receiver equivalently convert the system into an over-determined MIMO system, and the soft information update of the error symbols can be performed with simple linear receivers. Simulation results demonstrate that our proposed sparsity-aided iterative receiver can achieve significant performance gains over conventional UD-MIMO receivers, at the cost of a small complexity overhead.