An image retrieval method based on color histogram is faster than a retrieval method based on texture property in content-based retrieval. In this paper, we propose a multi-step retrieval method. For the first step, we can approximately classify using a histogram intersection technique from the image database. 또 the second step, we extract an image that classifies with detail using shape features related to images classified in the first step. As a result of the experiment, we have shown that our combined method shows improved performance in terms of retrieval effectiveness.