In multiple-input multiple-output (MIMO) systems Maximum Likelihood Detection (MLD) is known as an optimal method for signal detection in terms of bit error rate (BER). However, the main drawback of MLD is its high complexity. One of the possible methods to reduce the complexity of MLD is a signal detection scheme combining Zero-Forcing (ZF) and K-best algorithms based on the metric. The other method is the signal detection scheme with controlling the number of surviving candidates by setting the threshold to the metric. In this paper, we propose the signal detection scheme combining Minimum Mean Square Error (MMSE) Vertical Bell Laboratories Layered Space- Time (V-BLAST) and variable K-best algorithms in MIMO systems. In the proposed detection scheme, we estimate the initial symbol by MMSE V-BLAST algorithm and perform the variable K-best algorithm controlling the number of surviving candidates by setting the threshold to the metric. We show by simulation results that the proposed signal detection scheme can achieve almost identical BER to that of V-BLAST + K-best algorithm with less complexity and can achieve better BER than ZF + variable K-best algorithm with less complexity.