Knowledge distillation, which leverages efficientuni-modal models within their respective modalities,offers a promising strategy to reduce the sizeand enhance the efficiency of multi-modal models.However, addressing the challenges inherentto cross-modality—such as data heterogeneityand representation alignment—necessitates crossmodalknowledge distillation (CMKD) techniquesspecifically designed for diverse multimodal tasks.In this survey, we provide a comprehensive reviewof the key techniques and theoretical foundationsof CMKD. We explore its applications across variousmultimodal tasks, including video and audioprocessing, autonomous driving, human-centeredcomputing, medical analysis, and vision-languagemodels. Through this systematic analysis, weaim to offer valuable insights into the currentstate of CMKD research, its practical implementations,and potential future directions for advancingCMKD capabilities.