Signal recovery techniques for compressive sampling, such as matching pursuit (MP), orthogonal MP (OMP), and linear programming (LP), have been developed and applied to the spare channel estimation problem in orthogonal-frequency-division-multiplexed (OFDM) systems. In this paper, we first propose using a newly developed algorithm, subspace pursuit (SP), for the channel estimation problem. It is shown that the SP algorithm can outperform the MP, OMP, and LP algorithms. However, the SP algorithm will fail if the number of the pilots is small. We then propose a decision-feedback method, referred to as the pseudo-pilot assisted (PA) SP, which can be used in low pilot-density scenarios. Simulations show that the proposed PASP can significantly improve the performance of channel estimation when the number of pilots is small.