The paper introduces SemEval-2024 Task 3, titled "Multimodal Emotion Cause Analysis in Conversations," which aims to extract emotion and cause pairs from conversations. The task consists of two subtasks: Textual Emotion-Cause Pair Extraction in Conversations (TECPE) and Multimodal Emotion-Cause Pair Extraction in Conversations (MECPE). The dataset, ECF 2.0, is sourced from the sitcom *Friends* and contains 1,715 conversations and 16,720 utterances, with 12,256 emotion-cause pairs annotated at the utterance level across three modalities (language, audio, and vision). The evaluation metrics include F1 scores for each emotion category and a weighted average of F1 scores across six emotion categories. The competition attracted 143 registrations and 216 successful submissions, with 18 teams submitting system descriptions. The top-performing teams used advanced Large Language Models (LLMs) and various pipeline systems, demonstrating the importance of multimodal information and task-specific fine-tuning. The paper also discusses the challenges and future directions, including dataset bias, the utilization of LLMs, and the potential of multimodal information.The paper introduces SemEval-2024 Task 3, titled "Multimodal Emotion Cause Analysis in Conversations," which aims to extract emotion and cause pairs from conversations. The task consists of two subtasks: Textual Emotion-Cause Pair Extraction in Conversations (TECPE) and Multimodal Emotion-Cause Pair Extraction in Conversations (MECPE). The dataset, ECF 2.0, is sourced from the sitcom *Friends* and contains 1,715 conversations and 16,720 utterances, with 12,256 emotion-cause pairs annotated at the utterance level across three modalities (language, audio, and vision). The evaluation metrics include F1 scores for each emotion category and a weighted average of F1 scores across six emotion categories. The competition attracted 143 registrations and 216 successful submissions, with 18 teams submitting system descriptions. The top-performing teams used advanced Large Language Models (LLMs) and various pipeline systems, demonstrating the importance of multimodal information and task-specific fine-tuning. The paper also discusses the challenges and future directions, including dataset bias, the utilization of LLMs, and the potential of multimodal information.