We introduce Audio-Visual Contrastive Decoding (AVCD), a training-free framework for mitigating hallucinations in AV-LLMs by reformulating the existing contrastive decoding framework to support trimodal interactions.