In this paper, an iterative receiver for a joint data-detection and channel-estimation scheme is presented for orthogonal frequency-division multiplexing systems, which incorporates iterative decoding in the receiver. In the proposed scheme, a maximum a posteriori-based decoder and a channel estimator provide more reliable information on the coded bits for each other in an iterative manner. We first consider a practical implementation issue for the optimal minimum mean squared error two-dimensional (2-D) channel estimator as an essential element in the iterative receiver. To reduce the complexity of the 2-D estimator as suited to the iterative receiver, we focus on rigorously investigating how a separable estimator must be designed so that its structure may become asymptotically equivalent to that of the optimal 2-D estimator. Furthermore, we derive an analytical expression of the iterative process to evaluate a convergence performance as a function of the number of iterations and discuss its convergence property. Our simulation results demonstrate that the proposed iterative receiver achieves a near-ideal performance with only a few iterations under time-variant multipath fading channels.