In this paper, we propose a non-iterative symbol-wise beamforming scheme for MIMO-OFDM systems, which can provide the performance approaching that of the iterative symbol-wise beamforming scheme and can reduce the computational complexity significantly for channels with a small number of strong channel taps. Simulation results show that our proposed scheme leads to a negligible performance loss compared with the iterative symbol-wise beamforming scheme regardless of spatial correlation or presence of co-channel interference. For realistic situations, we also consider the system designs to manage imperfect channel knowledge at the transmitter or receiver. First, we propose a limited feedback technique and a codebook design algorithm for symbol-wise beamforming in a spatially correlated frequency selective channel. Simulation results show that the proposed limited feedback technique can provide insignificant outage capacity loss compared with the symbol-wise beamforming scheme with perfect channel knowledge when the proposed codebook is adopted with a moderate number of feedback bits. Second, we propose a robust transceiver optimization for symbol-wise beamforming under channel uncertainty. From simulation results, we confirm that our proposed robust transceiver optimization can incur less degradation than the iterative and proposed symbol-wise beamforming schemes under the assumption of a bounded channel uncertainty model.