It is a prevailing theoretical claim that path integration is the primary means of developing global spatial representations. However, this claim is at odds with reported difficulty to develop global spatial representations of a multiscale environment using path integration. The current study tested a new hypothesis that locally similar but globally misaligned rooms interfere with path integration. In an immersive virtual environment, participants learned objects' locations in one room and then physically walked, while being blindfolded, to a neighboring room for testing. These rooms were rectangular but globally misaligned. Adopting different actual perspectives in the testing room, the participants judged relative directions (JRDs) from the imagined perspectives in the learning room. The imagined and actual perspectives were aligned or misaligned according to either local room structures or global cardinal directions. Prior to JRDs, participants did not conduct other tasks (Experiment 1) or judged relative global headings of the two rooms to activate global representations while seeing the testing room (Experiment 2) or in darkness (Experiment 3). Participants performed better at locally aligned than misaligned imagined perspectives in all experiments. Better performances for globally aligned imagined perspectives appeared only in Experiment 3. These results suggest that structurally similar but misaligned rooms interfered with updating global heading by path integration, and this interference occurred during but not after the activation of global representations. These findings help to settle the inconsistency between the theoretical claims and empirical evidence of the importance of path integration in developing global spatial memories.