3D shape retrieval has gained popularity in recent years. Yet we still have difficulty in preparing a 3D shape by ourselves for query input. Therefore an easy way of doing 3D shape search is much awaited in terms of query input. In this paper, we propose a new method for defining a feature vector for 3D shape retrieval from a single 2D photo image. Our feature vector is defined as a combination of Zernike moments and HOG (Histogram of Oriented Gradients), where these features can be extracted from both a 2D image and a 3D shape model. Comparative experiments demonstrate that our approach exhibits effectiveness as an initial clue to searching for more relevant 3D shape models we have in mind.