It is known that activity in secondary motor areas during observation of human limbs performing actions is affected by the observer’s viewpoint, with first-person views generally leading to stronger activation. However, previous neuroimaging studies have displayed limbs in front of the observer, providing an offset view of the limbs without a truly first-person viewpoint. It is unknown to what extent these pseudo-first-person viewpoints have affected the results published to date. In this experiment, we used a horizontal two-dimensional mirrored display that places virtual limbs at the correct egocentric position relative to the observer. We compared subjects using the mirrored and conventional displays while recording over the premotor cortex with functional near-infrared spectroscopy. Subjects watched a first-person view of virtual arms grasping incoming balls on-screen; they were instructed to either imagine the virtual arm as their own [motor imagery during observation (MIO)] or to execute the movements [motor execution (ME)]. With repeated-measures anova, the hemoglobin difference as a direct index of cortical oxygenation revealed significant main effects of the factors hemisphere (P = 0.005) and condition (P ≤ 0.001) with significant post hoc differences between MIO-mirror and MIO-conventional (P = 0.024). These results suggest that the horizontal mirrored display provides a more accurate first-person view, enhancing subjects’ ability to perform motor imagery during observation. Our results may have implications for future experimental designs involving motor imagery, and may also have applications in video gaming and virtual reality therapy, such as for patients following stroke.