Abstract: Statistical pruning sphere decoder (SPSD) can effectively approximate the performance of maximum likelihood(ML) algorithm when using suitable pruning function, but its complexity is higher at low SNR. This paper focuses on lowering the complexity of SPSD and proposes an improved algorithm. In this algorithm, the initial radius, calculated by zero forcing(ZF) solution, can effectively reduce the search range and optimize the pruning function, which can lower the complexity effectively without losing performance. The simulation results show that for MIMO systems, the proposed algorithm can achieve similar performance to SPSD at low complexity such that the trade off between the performance and complexity of algorithm can be obtained.