A semi-blind channel estimation algorithm based on subspace approach for orthogonal frequency division multiplexing(OFDM) systems over the frequency-selective channel is proposed. A linear precoding is applied on each block before the IFFT operation and a low-rank structure is created in the received signal. Then subspace properties can be exploited to identify the channel up to a scalar ambiguity. The residual scalar ambiguities eliminated by inserting pilots into data stream. Simulation results illustrate the performance of the proposed semi-blind algorithm.