This paper presents a statistical approach for wavelet-based texture retrieval using generalized Gaussian density (GGD) and Kullback–Leibler distance (KLD). The method combines feature extraction (FE) and similarity measurement (SM) into a joint modeling and classification scheme. It uses GGD to model the marginal distribution of wavelet coefficients and computes KLD between estimated GGD models for similarity measurement. This approach is asymptotically optimal in terms of retrieval error probability. The method provides greater accuracy and flexibility in capturing texture information, while its simplified form resembles existing methods using energy distribution in the frequency domain. Experimental results on a database of 640 texture images show that the new method significantly improves retrieval rates, e.g., from 65% to 77%, compared with traditional approaches, while maintaining comparable computational complexity. The statistical framework is applied to wavelet-based texture retrieval, where wavelet coefficients in each subband are modeled by GGD. This results in a new texture similarity measurement in the wavelet domain with theoretical justification and no need for normalization. The method is shown to be effective in texture retrieval, with results demonstrating superior performance compared to traditional methods. The approach is also applicable to other retrieval methods and can be extended to more complex texture models. The paper concludes that the proposed statistical framework provides a justified way of defining similarity functions and can be used as a common framework for other existing methods.This paper presents a statistical approach for wavelet-based texture retrieval using generalized Gaussian density (GGD) and Kullback–Leibler distance (KLD). The method combines feature extraction (FE) and similarity measurement (SM) into a joint modeling and classification scheme. It uses GGD to model the marginal distribution of wavelet coefficients and computes KLD between estimated GGD models for similarity measurement. This approach is asymptotically optimal in terms of retrieval error probability. The method provides greater accuracy and flexibility in capturing texture information, while its simplified form resembles existing methods using energy distribution in the frequency domain. Experimental results on a database of 640 texture images show that the new method significantly improves retrieval rates, e.g., from 65% to 77%, compared with traditional approaches, while maintaining comparable computational complexity. The statistical framework is applied to wavelet-based texture retrieval, where wavelet coefficients in each subband are modeled by GGD. This results in a new texture similarity measurement in the wavelet domain with theoretical justification and no need for normalization. The method is shown to be effective in texture retrieval, with results demonstrating superior performance compared to traditional methods. The approach is also applicable to other retrieval methods and can be extended to more complex texture models. The paper concludes that the proposed statistical framework provides a justified way of defining similarity functions and can be used as a common framework for other existing methods.