In this paper, we introduce robust cyclic space- frequency (CSF) filtering for systems combining bit-interleaved coded modulation (BICM) and orthogonal frequency division multiplexing (OFDM). The proposed robust CSF filtering scheme exploits outdated channel state information at the transmitter (CSIT) and takes into account the reliability of the CSIT via a Bayesian model. Based on an upper bound on the worst-case pairwise error probability we formulate the optimization problem for the robust CSF filters which can be solved exactly for certain special cases. For the general case, we obtain an approximate solution by solving a related problem with additional constraints. This approximate solution can be further improved with a gradient algorithm for the original problem. Simulation results confirm the excellent performance of BICM-OFDM with robust CSF filtering.