This invention discloses a relative channel estimation method of MIMO based on self-adaptive training sequence. The process is that, the receptor determines the best length of the training sequence according to relative known information of the channel and transfers the value of the length and relative information of the channel through feedback link to the transmitter. The transmitter uses this feedback information to compute the optimal training sequence correspondent to current state of the channel according to the training sequence expression St=UD1/2tU*t designed by this model and transfers this optimal training sequence to the receptor through the forward link. Then the training cycle begins and the transmitter launches the training sequence to wireless channel. The receptor estimates the channel parameters of current time according to the known training sequence and the receipted signal in the training cycle and by using the minimum mean square error estimation criteria. The training sequence designed by this invention can make self-adaptive adjustment according to the needs of the actual system and the change of relative information of the channel. The invention has advantages of high estimation performance and strong robustness, thus can be used in wireless communication system with muti-antenna MIMO.