Natural image retrieval using low-level visual features is a challenging problem for content-based image retrieval. In this paper, a region-based image retrieval (RBIR) approach is proposed. Several feature vectors extracted from homogeneous color regions within an image represent each image, and similar images are retrieved based on these region features. In the experimental image database all images are grouped into 16 categories using moment feature to speed up the retrieval performance. Color mean, color histogram and moment of regions are used as features. From the experiments, it is found that a region-based retrieval returns more relevant images than using features based on the entire image.