With the development of online shopping, picture presentation as a shopping guide plays an increasingly important role in the online transaction. In many applications, it is very necessary to distinguish graphics and photographs before data mining. Based on the pictures of goods on the Internet, several ways are discussed to distinguish graphics from photographs, and their limits are analyzed. Then, a simple and effective algorithm is proposed. Firstly, text-images were divided from photographs by human visual habits, and token as graphics. Then, according to the analysis of dominant color and color diversity of the left images, the part generated by computer was figured out. The result on an online goods picture library including 40 000 pictures demonstrates the effectiveness and applicability of the algorithm.