We consider the iterative decoding of bit interleaved coded modulation (BICM) applied to both multiband and MIMO OFDM systems for typical scenarios where only a noisy (possibly very bad) estimate of the channel is provided by sending a limited number of pilot symbols. First, by using a Bayesian framework involving the channel a posteriori density, we propose a practical decoding metric that is robust to the presence of channel estimation errors. Then the derived metric is used in the demapping part of BICM multiband and MIMO OFDM receivers. We also compare our results with the performance of a mismatched decoder that replaces the channel by its estimate in the decoding metric. Numerical results over both realistic UWB and theoretical Rayleigh fading channels show that the proposed method provides significant gain in terms of bit error rate compared to the classical mismatched detector, without introducing any additional complexity.