The paper introduces EmoBench, a comprehensive benchmark designed to evaluate the emotional understanding capabilities of Multimodal Large Language Models (MLLLs). It also proposes EmoLLM, a novel model that incorporates two key techniques: Multi-perspective Visual Projection (MVP) and EmoPrompt. MVP captures diverse emotional cues from visual data by considering multiple viewpoints, while EmoPrompt guides MLLMs to reason about emotions in the correct direction. The benchmark includes a diverse dataset of approximately 287,000 multimodal instructions across various emotional tasks, such as emotion recognition and intent understanding. Experimental results show that EmoLLM significantly improves emotional understanding performance, with an average improvement of 12.1% across multiple foundation models. The work contributes to advancing MLLMs by enhancing their ability to understand and respond to complex human emotions, paving the way for applications in human-computer interaction, mental health support, and empathetic AI systems.The paper introduces EmoBench, a comprehensive benchmark designed to evaluate the emotional understanding capabilities of Multimodal Large Language Models (MLLLs). It also proposes EmoLLM, a novel model that incorporates two key techniques: Multi-perspective Visual Projection (MVP) and EmoPrompt. MVP captures diverse emotional cues from visual data by considering multiple viewpoints, while EmoPrompt guides MLLMs to reason about emotions in the correct direction. The benchmark includes a diverse dataset of approximately 287,000 multimodal instructions across various emotional tasks, such as emotion recognition and intent understanding. Experimental results show that EmoLLM significantly improves emotional understanding performance, with an average improvement of 12.1% across multiple foundation models. The work contributes to advancing MLLMs by enhancing their ability to understand and respond to complex human emotions, paving the way for applications in human-computer interaction, mental health support, and empathetic AI systems.