Low-light image enhancement (LLIE) is a challenging inverse problem due to information loss and noise disturbance in dark regions. Most existing methods overlook the frequency information in the Fourier domain or struggle to balance the enhancement performance and computational costs. In this paper, we empirically find that spatial information in the image and frequency information in the Fourier domains are essential and complementary for color recovery and noise removal. As such, we propose a Spatial-Frequency Combinatorial Network (SFCNet) for LLIE, exploring both spatial and frequency features simultaneously for effective enhancement. Specifically, we design a dual-branch spatial-frequency block (DSFB) that consists of a spatial branch and a frequency branch to model spatial information and frequency details in parallel, where the spatial branch extracts local spatial details, and the frequency branch captures global frequency dependency. Considering that the spatial and frequency features are within different modalities, we further introduce a cross-attention combinatorial interaction module (CCIM) to fuse local spatial and global frequency information mutually. Quantitative and qualitative experimental results show that the proposed method outperforms state-of-the-art methods.