Abstract: Existing code clone search approaches focus on the search of code clones with the same or similar implementation, but do not perform well for searching semantic code clones. To improve the accuracy of semantic code clone search, this paper proposes a code clone search approach based on code features. This approach builds code graph for each code fragment, then extracts key nodes to construct semantic feature representation from the graph, and uses inverted index and the TF-IDF scoring algorithm to search semantic code clones. Experimental results show that the proposed approach is better than the existing approaches in semantic code clone search.