A threshold based Schnorr-Euchner (TSE) sphere decoder (SD) is presented in this paper to reduce the high computational burden incur in multiple input multiple output (MIMO) decoding. The proposed method makes use of the probabilistic threshold for SE enumeration to remove the nodes which are most unlikely to be an ML solution. The threshold is chosen using the pruning probability based on noise statistics. By increasing the pruning probability, the higher level of complexity reduction can be achieved at the cost of slight performance loss. Simulations illustrate that the proposed TSE-SD achieves an improvement in performance through complexity reduction over conventional SE-SD.