In this paper, a feature extraction method suitable for describing local surface structure of the nonrigid 3D model is proposed, which is called histogram of oriented gradient based heat kernel signature (HOGHKS) extraction method. The method firstly extracts the heat kernel signature of the 3D point with isometric invariance, which makes the following extracted feature vector have isometric invariant characteristic and good stability. Then, the logarithm difference of the heat kernel signature is computed and its histogram of oriented gradient is computed, which can make the constructed feature vector have certain scale invariance to the scale variation of 3D model. The proposed feature in a certain extent solves the problems that the HKS feature does not have scale invariance, and the SIHKS feature has scale invariance though, it requires transforming the heat kernel signature into frequency domain for description, which will lose part of the effective description information. Extensive experiment results show that the HOGHKS feature has better retrieval performance compared with the HKS feature and SIHKS feature.