Abstract: an algorithm of channel estimation was proposed based on compressive sampling matching pursuit (CoSaMP) for multiple input multiple output-orthogonal frequency division multiplex (MIMO-OFDM) communication system, by exploiting the sparse characteristic of underwater acoustic channel. Firstly, the measurement matrix was used to establish a proxy for the channel, then matching pursuit idea was used to detect the position of the nonzero tap coefficients; finally, the least square algorithm was utilized to estimate the exact value of the nonzero coefficients. The channel estimation was finished through the estimation of the proxy. Comparing with conventional least square (LS) estimation algorithm, the channel estimation was greatly improved. Moreover, the algorithm outperforms Orthogonal Matching Pursuit (OMP) algorithms in computational complexity and estimate accuracy. The result of simulation has analyzed the performance and compared the channel estimation performance in different communication systems.