Abstract. In this paper, an interactive image retrieval scheme using MPEG-7 visual descriptors is proposed. The performance of image retrieval systems is still limited due to semantic gap, which is created from the discrepancies between the computed low-level features (color, texture, shape, etc.) and user's conception of an image. As a result, more interest has been created towards development of efficient learning mechanism other than designing sophisticated low-level feature extraction algorithms. A simple relevance feedback mechanism is proposed, that learns user's interest and updates feature weights based on a fuzzy feature evaluation measure. This has an advantage of handling comparatively small number of samples over those using standard classifiers involving large number of training samples and having more complexity. Extensive experiments have been performed to test to what extent the performance of an image retrieval system can be enhanced further using MPEG-7 standard visual features at minimum cost.Efficient image retrieval techniques from a large database have become an active field of research with the advent of the World-Wide Web. Content-Based Image Retrieval(CBIR) techniques are becoming more important with this basic requirement [1]. It is aimed at retrieving relevant images from an image database by measuring similarity between the automatically derived low-level features (color, texture, shape, etc. ) of the query image and the images stored in the database. Although different image characterization methods [2], [3] have been explored to represent images with basic low-level features but their usefulness is limited by the gap, between low-level features and high-level concepts known as semantic gap. Performance of CBIR is still far from user's expectations owing to semantic gap.To facilitate effective use of audio, visual(color, texture, shape, etc.) and motion descriptors, ISO/IEC has launched MPEG-7 to address multimedia retrieval. It provides a collection of specific, standard descriptors [4] used as a benchmark for evaluation of new schemes for image retrieval [5]. Among various state-of-the-art techniques in narrowing down the semantic gap, relevance feedback mechanism has been identified as an essential tool to provide significant performance boost in CBIR systems [6], [7], [8], through continuous learning and interaction with