Based on the analysis of color histogram for image retrieval, a new descriptor, bit-plane distribution entropy (BPDE), is proposed in this paper. The image is firstly divided into eight bit-planes and the Gray-code of bit-planes is introduced to avoid the effect of changes in the intensity values on bit-planes. Then, an entropy vector is constructed by computing the entropy of the first four significant planes which contain most of the structural information of the image. Finally, with designing of the correlation-weighted matrix, the Mahalanobis distance is adopted to measure the similarity because of the correlation between the concerned vectors. Comparisons are conducted between BPDE and other descriptors. Experimental results show that the proposed method provides more significantly retrieval results than the traditional ones.