RadarDistill is a novel knowledge distillation (KD) method designed to enhance the representation of radar data by leveraging LiDAR data. The method addresses the challenges posed by the noisy and sparse nature of radar data, which often leads to inefficient knowledge transfer from LiDAR to radar. RadarDistill employs three key components: Cross-Modality Alignment (CMA), Activation-based Feature Distillation (AFD), and Proposal-based Feature Distillation (PFD). CMA enhances the density of radar features using multiple layers of dilation operations, making it easier to align with the more densely distributed LiDAR features. AFD selectively transfers knowledge based on regions of LiDAR features, focusing on areas with high activation intensity. PFD guides the radar network to mimic LiDAR features within object proposals, improving the accuracy of object detection. The proposed method achieves state-of-the-art performance in radar-only object detection tasks, outperforming existing methods by significant margins in metrics such as mAP and NDS. Additionally, RadarDistill significantly improves the performance of camera-radar fusion models. The key contributions of RadarDistill include its ability to effectively transfer LiDAR knowledge to radar, the introduction of novel KD methods, and the substantial performance improvements in radar-based 3D object detection.RadarDistill is a novel knowledge distillation (KD) method designed to enhance the representation of radar data by leveraging LiDAR data. The method addresses the challenges posed by the noisy and sparse nature of radar data, which often leads to inefficient knowledge transfer from LiDAR to radar. RadarDistill employs three key components: Cross-Modality Alignment (CMA), Activation-based Feature Distillation (AFD), and Proposal-based Feature Distillation (PFD). CMA enhances the density of radar features using multiple layers of dilation operations, making it easier to align with the more densely distributed LiDAR features. AFD selectively transfers knowledge based on regions of LiDAR features, focusing on areas with high activation intensity. PFD guides the radar network to mimic LiDAR features within object proposals, improving the accuracy of object detection. The proposed method achieves state-of-the-art performance in radar-only object detection tasks, outperforming existing methods by significant margins in metrics such as mAP and NDS. Additionally, RadarDistill significantly improves the performance of camera-radar fusion models. The key contributions of RadarDistill include its ability to effectively transfer LiDAR knowledge to radar, the introduction of novel KD methods, and the substantial performance improvements in radar-based 3D object detection.