According to the threats of general leakage, cross leakage, similar leakage and multidimensional independent leakage in redistribution of data of multi sensitive attributes, this paper puts forward the concept of sensitive attribute sensitivity level and sensitivity level of sensitive attribute values. Then, it separates each dimension of sensitive attributes based on l-diversity model. It also puts forward the lq-diversity model of differentiated multi sensitive property. Experiments prove that it is safe for the distribution of data of multiple sensitive attributes. Finally, according to lq-diversity model, the paper puts forward a corresponding DMSA algorithm, which is proved to be correct and feasible, and has low occult rate and loss degree of additional information. The result indicates data has high availability and good privacy after released with this method.