Few-shot object detection (FSOD) aims to detect never-seen objects using few samples. Most meta-learning-based methods construct feature matching mechanisms between support samples and query proposals to facilitate object detection. However, the scarcity and varying quality of support samples, together with limited inter-class separability in the embedding space, often result in biased feature matching. To address these challenges, we propose a Superclass-guided Feature Enhancement framework for FSOD (SFE-FSOD) that mitigates feature matching bias by incorporating superclass through two key modules. First, a Superclass-guided Feature Enhancement (SFE) module is proposed, which strengthens category-relevant features within proposals using superclass prototypes to build more discriminative object representations. Second, a Superclass Group Contrastive (SGC) loss module is proposed to improve inter-class separability by applying differentiated contrast within superclass groups. Extensive experiments on the PASCAL VOC and MS COCO datasets demonstrate that the proposed SFE-FSOD achieves competitive performance in most evaluation scenarios, by reducing over-reliance on a few support samples during the feature matching process.