Semantic symmetry (e.g., “intelligent” vs. “stupid”) is a key structural feature of semantic information that enhances the flexibility of human semantic cognition and communication. Although previous neuroscience research has extensively investigated the neural representation of semantic information using various approaches, it is still unclear how symmetric semantic information is represented in the human brain. To uncover the representational patterns of symmetric semantic information in the human cerebral cortex using a data-driven approach, we conducted voxelwise modeling of movie-evoked cortical responses measured with functional magnetic resonance imaging (fMRI). Our voxelwise model predicted movie-evoked cortical responses from manual movie-scene annotations with thirty word labels that contained fifteen pairs of words with semantic symmetry (e.g., “human” and “mechanical”). This model enables us to identify the cortical locations of semantic representations associated with each word. The results revealed that the localization of semantic representations linked with each word manifested distributed patterns across the cortex. However, the representations for each symmetric word pair were rarely overlapped, and the overlap probability was significantly less than the chance level for all pairs. Rather, these representations were localized in highly asymmetric, heterogeneous patterns for each symmetric pair. These asymmetric and heterogeneous representations of semantic symmetry suggest the intricate nature of neural processes underlying semantic cognition. Our findings offer important evidence suggesting the complexity of neural mechanisms for human semantic cognition.