<p>Domain adaptation addresses the performance degradation of machine learning models caused by distributional differences between training and deployment environments, a challenge that is particularly critical in applications such as medical imaging, remote sensing, and object detection where labelled target-domain data is scarce or unavailable. In this context, contrastive learning has emerged as a powerful self-supervised paradigm for learning robust, transferable, and domain-invariant feature representations by leveraging instance discrimination and similarity-based objectives without relying on manual annotations. This review provides a comprehensive and methodologically grounded synthesis of contrastive learning approaches for domain adaptation, systematically analysing the literature across foundational principles, methodological advancements, and application-driven perspectives. We examine core components of contrastive learning, including positive–negative pair construction, network architectures, and contrastive loss functions, and categorize existing methods into instance-level, cross-domain, multi-source, adversarial, and hybrid frameworks. Particular emphasis is placed on contrastive learning for domain-adaptive object detection, especially in challenging scenarios involving small, dense, and occluded objects. Applications spanning medical imaging, autonomous driving, remote sensing, and natural scene understanding are reviewed to demonstrate the versatility of contrastive adaptation techniques. The review further identifies persistent challenges such as noisy negative samples, adversarial training instability, high computational cost, limited generalization to unseen domains, and the lack of standardized evaluation protocols, and discusses emerging research directions aimed at developing scalable, efficient, interpretable, and privacy-aware contrastive domain adaptation models.</p>