A major problem raised by a region-based image retrieval system is the proper description of regions for efficient and semantically meaningful retrieval. In this paper we present CoIRS a novel Cluster Oriented Image Retrieval System. A distinguishing aspect of CoIRS is its integration of a robust unsupervised learning for the detection of regions. The segmentation is based on local color and texture features that allows cluster- or region-based search. In addition, a privileged component is the constructing of Cluster Signatures (CS) that include, color, texture, and shape features of the clusters centroids. The system constructs Region Signatures (RS) as well which includes region based features such as the invariant moments, area, and eccentricity. Also, another distinctive feature of the system is Feature ranking. Three features were used for ranking the signatures, color, texture, or shape. CoIRS framework proved to provide successful retrieval results supported by precision estimation. The system is evaluated using a database of 2000 images composed of different categories of images.