Abstract: To efficiently find all frequent itemsets from large data sets, a frequent itemsets mining algorithm based on MapReduce and matrix is proposed. The algorithm includes the following points: Transaction database is transformed into matrix, then the matrix is divided into multiple submatrices by vertical partitioning method. According to MapReduce model, submatrices are assigned to the nodes of cluster, and the support of candidate itemsets is parallelly computed by the nodes. The amount of communication generated during the execution of the algorithm is small. Load balance among nodes is achieved. The performance of the algorithm is tested on Hadoop platform. Experimental results show that the algorithm has good speedup and scalability, which is suitable for mining frequent itemsets from large data sets.