As the digital technology advances with each passing day and the internet is evolving so quickly, the use of digital images is increasing on the demand. More information is showed in terms of digital patterns or images in our daily life. Besides retrieving image data from a given image database by context, we can alternatively do that by the image features we prescribed. This method is then called content-based image retrieval, CBIR. The wavelet transform possesses the power of multi-resolutional analysis for digital images. It¡¦s bands are mutually independent so that good results can often be obtained from partial analyses. Although wavelet transform is usually used for image compression and texture analysis, it has also many recent applications in the area of image retrieval. In this research, we propose the use of some new image roughness features to represent the variation of image textures. After an image is transformed on the wavelet, we collect the roughness features as well as wavelet energy features from each band. These features are then used to sort out desired images. We can show that the features as used in this work can be extracted even when the images are altered by some rotation, partial magnification or viewpoint changes.