For Web image retrieval, two basic methods can be used for representing and indexing Web images. One is based on the associate text around the Web images; and the other utilizes visual features of images, such as color, texture, shape, as the descriptions of Web images. However, those two methods are often applied independently in practice. In fact, both have their limitations to support Web image retrieval. This paper proposes a novel model called ’multiplied refinement’, which is more applicable to combination of those two basic methods. Our experiments compare three integration models, including multiplied refinement model, linear refinement model and expansion model, and show that the proposed model yields very good performance.