Low-light image enhancement (LLIE) has achieved remarkable progress as a hot research topic in the field of image restoration. However, existing methods lack explicit modeling of the coupling between frequency components and physical illumination properties, resulting in a lack of effective physical constraints across frequency bands. Consequently, it is difficult to achieve a good balance between enhancement strength and texture reconstruction. To address these issues, we propose an LLIE method based on a Retinex-Guided Wavelet Diffusion model, named RGWD. Our RGWD decomposes the low-light image into high-frequency and low-frequency components in the wavelet domain, and performs collaborative optimization of frequency-domain analysis and illumination correction under the guidance of Retinex knowledge, thereby simultaneously enhancing structural details and brightness. Specifically, we utilize the Retinex Decomposition Guidance Sub-network (RDGNet) to perform Retinex decomposition on the low-light image, obtaining the reflectance map that represents the physical properties of the scene, and the illumination map that describes the environmental lighting conditions, and further extract intrinsic scene features. To guide the generation of low-frequency components obtained by wavelet decomposition, we embed an Adaptive Selection and Fusion (ASF) module into the diffusion model, which imposes physical constraints on the diffusion process using Retinex enhanced and consolidated features. For high-frequency restoration, we utilize a High-Frequency Compensation (HFC) module, which leverages the inherent high-frequency information from the reflectance map to effectively supplement the texture details in low-light images. Extensive experimental results demonstrate the rationality and superior performance of our method.