The　problem　of　＇tied　trees＇　has　been　noticed　in　the　traditional　NJ　(Neighbor　Joining)　method,　which　may　produce　more　than　one　phylogenetic　tree　from　a　single　data　set,　depending　on　the　order　of　data　entry.　To　solve　this　problem,　this　paper　presented　an　improved　NJ　algorithm.　Being　different　from　the　traditional　NJ　algorithm,　the　improved　NJ　repeatedly　merged　multiple　groups　into　one　by　rate-corrected　distance　until　only　one　group　remained.　The　author　shows　in　theory　and　application　that　the　improved　algorithm　can　always　produce　a　unique　phylogenetic　tree,　which　is　the　same　as　the　original　NJ　tree　without　considering　the　order　of　branches　in　cases　that　they　are　actually　bifurcating.　Thus　it　may　produce　a　multifurcating　tree　with　a　unique　topology,　even　if　the　original　NJ　trees　are　tied.