Phylogenetic　tree　is　essential　to　understand　evolution　and　it　is　usually　constructed　through　multiple　sequence　alignment,　which　suffers　from　heavy　computational　burdens　and　requires　sophisticated　parameter　tuning.　Recently,　alignment　free　methods　based　on　k-mer　profiles　or　common　substrings　provide　alternative　ways　to　construct　phylogenetic　trees.　However,　most　of　these　methods　ignore　the　global　similarities　between　sequences　or　some　specific　valuable　features,　e.g.,　frequent　patterns　overall　datasets.　To　make　further　improvement,　we　propose　an　alignment　free　algorithm　based　on　sequential　pattern　mining,　where　each　sequence　is　converted　into　a　binary　representation　of　sequential　patterns　among　sequences.　The　phylogenetic　tree　is　further　constructed　via　clustering　distance　matrix　which　is　calculated　from　pattern　vectors.　To　increase　accuracy　for　highly　divergent　sequences,　we　consider　pattern　weight　and　filtering　redundancy　sub-patterns.　Both　simulated　and　real　data　demonstrates　our　method　outperform　other　alignment　free　methods,　especially　for　large　sequence　set　with　low　similarity.