Drug repositioning offers a cost-effective strategy for identifying novel therapeutic applications of existing drugs. However, the inherent complexity and heterogeneity of biomedical data pose significant challenges to accurate drug–disease association prediction. In this study, we propose HGAT-CL, a new framework that integrates Heterogeneous Graph Attention Networks (HGAT) with Contrastive Learning (CL) to enhance predictive performance in this domain. HGAT-CL constructs a heterogeneous graph based on drug–drug similarities, disease–disease similarities and known drug–disease associations. The model employs a dual-view encoding mechanism: (i) an intra-view encoder utilizing type-specific GATConv layers to independently learn from each edge type, and (ii) an inter-view encoder leveraging a HeteroGAT layer to capture global semantic interactions across the entire graph. These two representations are aligned via an InfoNCE-based contrastive loss, promoting consistency and discriminative feature learning. Empirical evaluations on the benchmark Cdataset demonstrate that HGAT-CL consistently outperforms state-of-the-art methods in terms of AUC and AUPR, highlighting its effectiveness for drug–disease association prediction.