Abstract: Due to the different effects of different features on multidimensional data during the clustering, this paper proposes an affinity propagation clustering algorithm based on adaptive feature weight. This method introduces a feature weight similarity measure to affinity propagation by considering different importance of different features. During the clustering the new method evaluates the objective function according to current data partition, and calculates the contribution of each features made to current clustering. Then the feature weights will be self-adaptive updated according to the contribution, and the two messages passing between data points in affinity propagation will be recalculated on the basis of feature weight similarity matrix to enhance the clustering results. The experimental results show that feature updating works well on the new algorithm, and compared to the affinity propagation and other traditional clustering algorithm the new algorithm can obtain better results.