Abstract: Channel estimation is one of the key techniques in wireless communication systems, its accuracy has a significant impact on the overall system performance. This paper proposes a subspace based blind channel estimation algorithm for Multi-Input Multi-Output with cyclic prefix (MIMO-CP) systems, which utilizes the redundancy introduced by the cyclic prefix. The channels can be estimated by taking singular-value decomposition to the auto-correlation matrix of received blocks. It is essential to get an accurate the estimation of auto-correlation matrix by using the received blocks as more as possible. However, too many received blocks may cause a long detection delay, which would not be used in the real-time signal processing systems, and moreover, it is difficult to track the channels varying with rapid speeds. To overcome the disadvantages mentioned above, we proposed a subspace based blind channel estimation algorithm by using fewer blocks in the MIMO-CP systems, thus this method is very suitable for blind estimation of the fast varying channels. According to the unique property of the MIMO-CP systems, which the channel matrix is a block circulant channel matrix, for the vector composed of the two consecutive received data blocks and the corresponding cyclic prefix part, we can obtain a group of new vectors without changing the system matrix. By utilizing these extended vectors, we can get an accurate the estimation of auto-correlation matrix with fewer blocks. We discuss the complexity of the proposed method, which is proportional to the length of block, the channel order and repetition times. Considering the whole system performances, the increasing of the complexity of our method is acceptable. This paper analyzes the system performance in different repetition factors and the number of received blocks. Simulation results show that there is a compromising between the repetition times and the needed received blocks. Furthermore, compared with some existing methods, such as “precoding” and “virtual carriers” blind channel algorithms, the proposed algorithm archived a more reliable estimation and better bit error rate performance with fewer received blocks.