In this paper, an expectation maximization (EM) algorithm based iterative receiver is utilized in the underwater acoustic OFDM system to improve the spectrum efficiency. A relatively smaller number of pilots are used to determine the initial value for the EM-based channel estimation process. The intermediate estimates of the channel state information generated by the EM algorithm are fed forward to the BCJR decoding unit to achieve the a posterior probability of the transmitted bit sequence. The a posterior probability is then fed back to the EM algorithm to estimate channel state information. The iterative receive process is expected to realize a better channel estimate and bit error rate (BER) performance. Simulation results demonstrate that the EM-based iterative underwater acoustic OFDM system has a fast convergence property, and can achieve a low mean square error (MSE) for the channel estimation and a near optimal BER performance.