Abstract: Clustering is an effective method for discovering the potential information in the fields of data mining. Aiming at the problem of traditional affinity propagation (AP) clustering algorithm denoted by Euclidean measure can not deal with the complicated data sets, a novel algorithm, affinity propagation with density similarity measurement (APDSM), is presented. Firstly, the idea of density is introduced. Then under the frame of traditional affinity propagation, the density gene is defined and novel similar-Euclidean measure is designed. A density sensitive similarity measurement is constructed as well. Finally experiment is used to validate the algorithm.