Channel estimation is required to demodulate the multiple-input multiple-output (MIMO) signals and there are blind and non-blind channel estimation schemes. One of non-blind channel estimation schemes is using Gibbs sampling. In channel estimation using Gibbs sampling, when the number of received data symbols is small, estimation accuracy may be deteriorated owing to the unreliability of weights. In order to solve this problem, this paper introduces a coefficient for adjusting the weights used in channel estimation with Gibbs sampling. When the number of data symbols is small, larger coefficient values are found to be more suitable because they suppress the fluctuation of channel estimation. Numerical results obtained though computer simulation show that the proposed scheme improves by about 5dB as compared with a decision directed scheme in a 4 × 4 MIMO system when the number of data symbols is one.