Density peaks clustering algorithm (DPC) relies on local-density and relative-distance of dataset to find cluster centers. However, the calculation of these attributes is based on Euclidean distance simply, and DPC is not satisfactory when dataset’s density is uneven or dimension is higher. In addition, parameter $$ d_{\text{c}} $$ only considers the global distribution of the dataset, a little change of $$ d_{\text{c}} $$ has a great influence on small-scale dataset clustering. Aiming at these drawbacks, this paper proposes a mass-based density peaks clustering algorithm (MDPC). MDPC introduces a mass-based similarity measure method to calculate the new similarity matrix. After that, K-nearest neighbour information of the data is obtained according to the new similarity matrix, and then MDPC redefines the local density based on the K-nearest neighbour information. Experimental results show that MDPC is superior to DPC, and satisfied on datasets with uneven density and higher dimensions, which also avoids the influence of $$ d_{\text{c}} $$ on the small-scale datasets.