This paper proposes a new iterative channel estimation algorithm for known symbol padding (KSP) Orthogonal Frequency Division Multiplexing (OFDM) based on the Expectation Maximization (EM) algorithm. The guard interval is filled with pilot symbols and the remaining part of the pilot symbols is put on some of the OFDM carriers. To start up the EM algorithm an initial channel estimate is obtained by using only the pilot symbols. Then the EM algorithm is applied until convergence is reached. The performance of this estimator is compared with the performance of the iterative joint data-aided (DA) / decision-directed (DD) estimator proposed in [1]. The EM algorithm converges to the performance of the all pilot estimator for lower SNR than the iterative joint DA/DD estimator, but the gain in performance results in a higher computational complexity.