Content-Based Image Retrieval (CBIR) is an automatic process of retrieving images that are the most similar to a query image based on their visual content such as colour and texture features. However, CBIR faces the technical challenge known as the semantic gap between high level conceptual meaning and the low-level image based features. This paper presents a new method that addresses the semantic gap issue by exploiting cluster shapes. The method first extracts local colours and textures using Discrete Cosine Transform (DCT) coefficients. The Expectation-Maximization Gaussian Mixture Model (EM/GMM) clustering algorithm is then applied to the local feature vectors to obtain clusters of various shapes. To compare dissimilarity between two images, the method uses a dissimilarity measure based on the principle of Kullback-Leibler divergence to compare pair-wise dissimilarity of cluster shapes. The paper further investigates two respective scenarios when the number of clusters is fixed and adaptively determined according to cluster quality. Experiments are conducted on publicly available WANG and Caltech6 databases. The results demonstrate that the proposed retrieval mechanism based on cluster shapes increases the image discrimination, and when the number of clusters is fixed to a large number, the precision of image retrieval is better than that when the relatively small number of clusters is adaptively determined.