Landmark-based and self-motion-based navigation are two fundamental forms of spatial navigation, which involve distinct cognitive mechanisms. A critical question is whether these two navigation modes invoke common or distinct spatial representations for a given environment in the brain. While a number of electrophysiological studies in non-human animals have investigated this question but yielded inconsistent results, it still awaits rigorous investigation in humans. In the current study, we combined ultra-high field fMRI at 7T and desktop virtual reality with state-of-the-art fMRI data analysis techniques. Using a novel linear track navigation task, we dissociated the use of landmarks and self-motion cues, so that participants used different spatial cues to encode and retrieve the same set of spatial locations. Focusing on the retrosplenial cortex (RSC) and the hippocampus, we observed that RSC contained both cue-specific and cue-independent spatial representations, which were driven by objective location (where the participant was actually located) and subjective location (the participant’s self-reported location), respectively. The hippocampus showed strong functional coupling with RSC and exhibited a similar spatial coding scheme, but with reduced effect sizes. Taken together, the current study demonstrated for the first time concurrent cue-specific and cue-independent spatial representations in RSC in the same spatial context, suggesting that this area might transform cue-specific spatial inputs into coherent cue-independent spatial representations to guide navigation behavior.