Abstract: An improved Apriori algorithm which uses support count matrix to generate frequent itemsets is proposed.It avoids the generation of a large amount of non-frequent candidate k-itemsets in the join step and the judgments of(k-1)-subset in the prune step by the characteristics of rows and columns in the upper triangular matrix which is partitioned and sparse,and by the restrict relations between non-frequent columns.It effectively compresses the search space and reduces the computational cost of the Apriori algorithm.