Lattice reduction-aided ordered successive interference cancellation (LRA-OSIC) detection is capable of achieving optimal diversity orders for multiple-input multiple-output (MIMO) communications. When the number of antennas is large, however, there can still be a significant gap between the performance achievable with the LRA-OSIC detector and the maximum likelihood detector (MLD). This paper introduces a regularization approach to enhance the performance of LRA-OSIC detectors. Multiple approximate models for the same MIMO channel are generated and a standard LRA-OSIC detector is then constructed for each model. The best detector is determined for each instantaneous received symbol, using a residual-based method. The search can be terminated using a stopping criterion. Simulation results show that significant performance enhancements can be achieved by the proposed design at only a moderate increase of complexity1.