Research on knowledge graph (KG) link prediction has primarily focused on single graphs. However, real-world knowledge is often distributed across multiple heterogeneous KGs, and the incompleteness of individual graphs limits their practical utility. In this paper, we study the emerging task of cross-KG link prediction—answering complex queries that require cross-graph information—and propose RuleGF, a Rule-Guided Graph Fusion model. We are the first to systematically introduce rule-based reasoning into this task, presenting an innovative framework that integrates rule reasoning with iterative graph fusion. By leveraging a link prediction-assisted filtering mechanism for entity alignment (EA), we establish a mutually reinforcing process between the two tasks. Furthermore, we design an iterative graph fusion framework with a conflict resolution strategy to progressively improve the quality of cross-graph fusion. Experiments on DBP-FB and WIKI-YAGO benchmarks demonstrate that RuleGF achieves state-of-the-art accuracy, with strong performance in computational efficiency and interpretability, offering a practical solution for cross-graph reasoning.