Knowledge Distillation (KD) aims to transfer knowledge from a cumbersome teacher network to a more compact student network, enabling efficient deployment on devices with constrained resources. Traditional KD methods rely on balanced datasets to ensure the effectiveness of model training. Nevertheless, in real-life situations, data often show a distribution with a long tail. Due to the limited number of tail samples, current KD methods frequently struggle to obtain sufficient information from these samples, impacting the overall performance. To tackle the challenge of insufficient tail class information, this work introduces a Dynamic Inter-class Correlation Knowledge Distillation(DIC-KD) method. Specifically, we first define dynamic inter-class correlations and extract rich inter-class knowledge from the teacher model. Then, we transfer this knowledge to the student model to enhance its learning ability for various categories, especially for tail class samples, thus improving the overall performance. Intensive experiments demonstrate that DIC-KD on long-tailed datasets can proficiently train stable student networks.