This paper introduces EVEDIT, an event-based knowledge editing framework that addresses the limitations of existing knowledge editing methods by incorporating event descriptions to define deduction anchors and editing boundaries. The key contributions include identifying the critical issue of improper deduction anchor assignments in current knowledge editing settings, proposing event-based knowledge editing and a new benchmark EVEDIT, and introducing a novel Self-Edit approach that significantly improves factual consistency and naturalness in edited models. The paper also highlights the challenges of existing methods in handling uncertain editing boundaries and demonstrates that event-based editing reduces uncertainty compared to traditional factual editing. The proposed framework is evaluated on both text completion and question-answering tasks, showing that Self-Edit outperforms existing methods in terms of factual consistency while maintaining the naturalness of generated text. The study emphasizes the importance of defining clear editing boundaries and the role of event-based descriptions in improving the logical soundness and practical applicability of knowledge editing in real-world scenarios.This paper introduces EVEDIT, an event-based knowledge editing framework that addresses the limitations of existing knowledge editing methods by incorporating event descriptions to define deduction anchors and editing boundaries. The key contributions include identifying the critical issue of improper deduction anchor assignments in current knowledge editing settings, proposing event-based knowledge editing and a new benchmark EVEDIT, and introducing a novel Self-Edit approach that significantly improves factual consistency and naturalness in edited models. The paper also highlights the challenges of existing methods in handling uncertain editing boundaries and demonstrates that event-based editing reduces uncertainty compared to traditional factual editing. The proposed framework is evaluated on both text completion and question-answering tasks, showing that Self-Edit outperforms existing methods in terms of factual consistency while maintaining the naturalness of generated text. The study emphasizes the importance of defining clear editing boundaries and the role of event-based descriptions in improving the logical soundness and practical applicability of knowledge editing in real-world scenarios.