This paper proposed a new method for extracting visualness attributes (the extent to which an attribute can be perceived visually) that based on conceptual graphs (CGs). By providing a small scale seed attributes, this method acquire the context which contain these seed attributes by two steps, primary entity matching and sentence selection, then transform the selected sentences into CG templates, after systematic expansion of its semantic information on the basis of HowNet lexicon, extract the attribute concepts by computing the similarity between CG templates and textual CGs, then compute the visualness of these attribute concepts and retain the attributes with the visualness value greater than the threshold. At last, we construct the relationship among the attributes by bringing in world knowledge. Experiments have demonstrated the effectiveness of our conceptual graph based method when compared with the state of art ones.