Abstract: This paper proposes a new algorithm for mining multi-dimensional sequential patterns.The algorithm mines sequential patterns in dataset firstly，and then finds frequent 1-itemset from multi-dimensional information that support this pattern in dataset for every sequence pattern，and generates frequent k-itemset from frequent （k-1）-itemset.This algorithm gets the frequent k-itemset through scanning the frequent （k-1）-itemset that reduce gradually，so it saves time.Experiment shows this method has good efficiency.