Sphere decoding (SD) for multiple-input multiple-output (MIMO) systems has been shown to achieve maximum-likelihood (ML) performance with low complexity. However, the achievement of such an excellent performance-complexity tradeoff is highly dependent on initial choice of sphere radius. In this paper, we propose a new scheme of initial radius selection which has less computational complexity than the original SD. Moreover, the new expanding method of sphere radius that we propose can detect symbols efficiently. In our simulations, new method can reduce computational complexity approximately 35 percent compared with original SD when SNR is smaller than 10dB.