In this paper we present a probabilistic framework for shape-based indexing and retrieval of images. In our framework shape-based features are extracted from each image and then a statistical model of the image is constructed using an effective determininstic method for Gaussian mixture modeling. In this way, each image is finally represented as a mixture of Gaussians and shape-based similarity between images is computed by measuring the distance between the corresponding mixture distributions. Several distance measures are presented and experimentally compared. Experimental results on the retrieval of logo images indicate that the method is very effective and exhibits robustness to the presence of various types of edge-related noise in the query image.