The acquisition of accurate channel estimation is important for Orthogonal Frequency-Division Multiplexing (OFDM) system to reliably recover the transmitted data symbols. Decision directed channel estimation (DDCE) has been utilized to improve the accuracy of conventional training-based channel estimation. However, this method can only provide limited improvement when the data decision feedback comprises a large portion of decision errors at low signal-to-noise (SNR) range. A new Iterative DDCE (IDDCE) scheme with reliable data feedback selection is proposed in this paper. The unreliable data decisions are eliminated on the subcarriers where noise components are enhanced due to the impact of frequency selective channel. An explicit expression of the variance of channel estimation error is derived, based on which the optimal selection criteria are also derived to minimize the variance. Simulation results show that the accuracy of channel estimation is significantly improved with the proposed algorithm.