Natural　scene　classification　is　a　challenging　pattern　classification　problem　nowadays.　The　description　of　image　plays　a　crucial　role　in　the　process　of　recognition.　Many　different　approaches　and　feature　extraction　methodologies　concerning　scene　classification　have　been　proposed　and　applied　in　the　last　few　years.　This　paper　proposed　a　novel　method　of　natural　scene　recognition　based　on　graph　edit　distance　(GED)　in　which　scene　images　are　represented　by　attributed　graph.　The　vertex　label　is　the　features　of　regions　and　edge　label　is　the　features　of　public　area　of　adjacent　regions.　This　method　used　local　representation　as　well　as　global　way,　realized　the　cooperation　of　global　and　local　mechanisms.　The　proposed　method　approaches　satisfactory　categorization　performances　on　the　well-known　scene　classification　datasets　with　8　scene　categories.　©　(2014)　Trans　Tech　Publications,　Switzerland.