Feature descriptors are the key factors influencing the result of non-rigid 3D model correspondence.But a single feature descriptor only contains one aspect of information of a 3D model.In order to overcome the limitation of single feature and further improve the accuracy of model correspondence, the entropy was introduced to calculate the weight of each single feature according to its correspondence results.The features of HKS(heat kernel signature), WKS(wave kernel signature) and surface area were fused with these weights.The effectiveness of the approach was evaluated by using the SHREC'2014 non-rigid 3D human models benchmark.In addition, the results outperform those of any state-of-the-art single feature descriptor, and can be used for non-rigid 3D model retrieval.