A novel approach to visually represent perceptual features of textures is proposed and analysed. This approach combines statistical properties of the texture together with qualitative measures coming from human perception. Quantitative measures, coming from co-occurrence analysis, have been in turn visually represented using an effective iconic representation, producing an immediate and easy tool to discriminate a texture from a larger set. Such a representation is very effective to formulate approximate queries to a database of textures for similarity based retrieval.