Most of the existing multiple-input multiple-output orthogonal frequency division multiplexing (MIMO-OFDM) channel estimation methods do not take into account the effect of the pulse-shaping filter in the transmitter nor of the matched filter in the receiver, thus leading to an estimation solution for the composite channel including the pulse-shaping and matched filters, instead of the pure wireless channel. This solution is not sufficiently accurate because of the extra length of the composite channel induced by the two filters especially in the scenario with a small pure channel length. In this study, the authors present a novel methodology for the estimation of the pure multipath channels of pulse-shaped MIMO-OFDM systems. By utilising the knowledge of pulse-shaping and matched filters, the authors develop two channel estimation approaches, namely, a semi-blind approach for the sampling duration-based channels, in which the multipath occurs at the sampling instant and a training-based least-square technique for the upsampling duration-based channels where the multipath may occur in a fraction of sampling duration. A number of computer simulation-based experiments are conducted, and these simulation results confirm the efficacy of the proposed approaches.