Hashing techniques are widely adopted in cross-modal retrieval owing to low memory consumption and computational cost. However, there are two issues: (1) The semantic correlations implied in modal features are not fully explored; (2) The complementarity between modality-specific semantics and label semantics is not adequately utilized. To address these problems, we propose a method called Multimodal Pseudo-label Guided Semantic Enhanced Hashing Learning for Cross-Modal Retrieval (MPGSE). The method consists of three main steps: (1) Constructing pseudo-labels for each modality by using fuzzy clustering to capture semantic correlations among modal features. (2) Exploring the complementarity between modality-specific semantics and label semantics, while reducing semantic differences between modalities through consensus learning. (3) Generating discriminative hash codes for each modality. Comprehensive experiments on two benchmark datasets fully demonstrate the excellent performance of MPGSE. The relevant code can be accessed at https://github.com/WuhahaWU/whh_code