Recently, multiple-input multiple-output (MIMO) systems that realize a high bit rate data transmission with multiple antennas at both transmitter and receiver have drawn much attention for high spectral efficiencies. In MIMO systems, eigen-beam space division multiplexing (E-SDM) that achieves good performance by weighting at the transmitter using channel state information (CSI) has been studied. Early studies for ESDM have assumed perfect CSI at the transmitter. However, in practice, CSI fed back to the transmitter is not identical to that when the signals are transmitted owing to the time-varying nature of channels and feedback delay. As a result, the performance of E-SDM is degraded. In this paper, as methods that reduce the performance degradation of E-SDM in the presence of feedback delay, we evaluate the performance of a method that predicts CSI when the signals are transmitted at the receiver and feeds the predicted CSI back to the transmitter (denoted by channel prediction method). We also evaluate the performance of a method that uses the receive weight based on zero-forcing (ZF) or minimum mean square error (MMSE) criterion instead of those based on singular value decomposition (SVD) criterion (denoted by ZF or MMSE receive weight method). Simulation results show that bit error rate (BER) degradation of E-SDM in the presence of feedback delay is reduced by three methods. We also show that the ZF and MMSE receive weight methods achieve the identical BER with smaller amount of calculation than the channel prediction method.