The paper presents a statistical approach to texture retrieval in content-based image retrieval (CBIR) systems, combining feature extraction (FE) and similarity measurement (SM) into a joint modeling and classification scheme. The authors propose using a generalized Gaussian density (GGD) to model the marginal distribution of wavelet coefficients and computing the Kullback–Leibler (KLD) distance between estimated GGDs for SM. This method is shown to be asymptotically optimal in terms of retrieval error probability. Experimental results on a database of 640 texture images indicate that the proposed method significantly improves retrieval rates, from 65% to 77%, compared to traditional approaches, while maintaining comparable computational complexity. The statistical framework is flexible and can be applied to other features and more general image retrieval systems.The paper presents a statistical approach to texture retrieval in content-based image retrieval (CBIR) systems, combining feature extraction (FE) and similarity measurement (SM) into a joint modeling and classification scheme. The authors propose using a generalized Gaussian density (GGD) to model the marginal distribution of wavelet coefficients and computing the Kullback–Leibler (KLD) distance between estimated GGDs for SM. This method is shown to be asymptotically optimal in terms of retrieval error probability. Experimental results on a database of 640 texture images indicate that the proposed method significantly improves retrieval rates, from 65% to 77%, compared to traditional approaches, while maintaining comparable computational complexity. The statistical framework is flexible and can be applied to other features and more general image retrieval systems.