Abstract: A 3D models retrieval method based on collaboration semantic annotation of discrepant feature is proposed.The method makes use of advantage of several feature extraction algorithms,and refers collaborative training approach in semi-supervised learning.Firstly,three classifiers by are trained with different feature extraction algorithms respectively,and three different classifiers are constructed.Secondly,3D models are semantic annotated with semantics automatically by the optimal classifier which was found during the course of collaborative iterative training.Finally,3D models retrieval with semantics is carried out.Experiments on PSB show that the annotation accuracy of this method is higher than the method based on single-feature.Meanwhile,the recall-precision rises.