Distance measurement is one of the key tasks in content-based image retrieval (CBIR). This paper proposes a new fractional distance metric for CBIR. We conduct extensive experiments on three famous benchmark datasets, using different color, texture and shape features. Our experiments show that retrieval performance of the new distance metric consistently outperforms the more common City Block and Euclidean distance metrics, as well as several other commonly-used distance functions. The experimental results on three commonly used benchmark datasets show that the new fractional distance metric can be used universally.