Light field data provide rich visual cues for saliency detection tasks through their multi-view information. Although depth maps can enhance the understanding of the 3D structure of a scene, they face challenges in spatial alignment with all-focus images and focal stacks. This paper proposes a depth-guided multimodal fusion network (DGMFNet) to overcome this limitation and fully utilize depth information. Specifically, we design a weight-guided fusion module (WGFM) and cross-modal fusion module (CMFM), where the WGFM is used to attenuate the difference between the depth feature and the other two features, and the CMFM is used to capture the depth information of the depth features by performing the enhanced features by combining them to capture deeper feature information. In addition, we introduce a shallow feature embedding module (SFEM) on the decoding side to improve the edge details of the salient map by extracting the shallow features of all-focus images. Experimental results show that our method outperforms several other methods on two extensive light field datasets.