Efficient retrieval of information from image and other multimedia databases if an important and challenging problem. Among techniques developed for this purpose, query by example is the most effective way for searching large collections of images. In this paper we propose a new approach to extracting image features based on the local intensity pattern. The user can select any region of interest within an image and request that the system retrieve all similar looking patterns. No domain specific or context information is used in analysis, and the feature representation has desirable properties such as reduced dimensionality, preserving perceptual similarity and offering a natural hierarchical indexing structure. Experimental results are presented.