Aiming at solving the representation and query efficiency in massive and dynamic dataset on storage system, a Multi-group Balance Matrix Bloom Filter (M-BMBF) and the algorithms on insertion and searching of data element were proposed. M-BMBF initiates a r×m matrix Bloom filter according to the size of dataset, and it introduces multiple located hash functions which can be used to divide the matrix Bloom filter into multi-group to achieve balanced insertion and efficient query operations. In order to slow down the bits consumption rate in Bloom filter when a new element is inserted, a longest-bit match filling algorithm was proposed, which selects a Bloom filter as the destination position for insertion from the candidate Bloom filters according to the rule that fewest bits will be changed due to this insertion operation. Experiment results show that compared with the classical Split Bloom Filter, M-BMBF can efficiently save storage space and decrease the misjudgment rate, while its time consume is constant.