A subspace approach for blind channel estimation in SIMO-OFDM systems without cyclic prefix is proposed in this paper. By exploiting a repetition index in the signal model, our algorithm generalizes the conventional subspace approach. The computational complexity of the singular value decomposition (SVD) operation in the proposed method is lower than that in other subspace approaches. Additionally, the proposed approach is suitable for cyclic-prefix-free systems. An extra benefit is that fewer OFDM blocks are needed to satisfy the necessary condition of the signal matrix to be full rank. Simulation results show that the proposed method is better than existing approaches in normalized mean square error.