INTRODUCTION: Scene-selective regions were shown to be significantly affected by spatial frequencies (SF) and have different sensitivities to low spatial frequencies (LSF) and high spatial frequencies (HSF). However, previous studies mainly focused on the neural activations or the neural patterns in a single SF band. METHODS: To investigate the extent to which the information of a single SF is used in scene category representations, we not only decoded the scene categories in each SF, but also used the neural patterns to LSF or HSF to decode the patterns to non-filtered (NF) scenes based on fMRI data using multivoxel pattern analysis (MVPA). As luminance contrast was shown to follow statistical regularities along with SF, we performed the decoding analyses separately in two conditions of contrast where the contrast of LSF and HSF was unmodified or equalized. RESULTS: The results showed distinct SF preferences in the two contrast conditions, showing that luminance contrast has a significant role in SF processing. In addition, we also performed the above analyses only within natural and indoor scenes, respectively. The results showed the scene-selective regions were more efficient in distinguishing natural scene categories in LSF, and the LSF was preferentially used along with high luminance contrast in recognition of natural scenes. On the other hand, humans preferentially used HSF information in distinguishing indoor scenes. DISCUSSION: This distinct SF preferences maybe caused by the different aspects of information conveyed by LSF and HSF, as well as the different strategies of spatial perception in natural and indoor scenes recognition.