In this paper we address two aspects of image retrieval. First, we present the retrieval of an object or a scene in the presence of important scale changes. The approach is based on the detection of scale invariant interest points. These points are used to characterize the image ; the scale associated with each point allows to compute scale invariant descriptors. Our descriptors are, in addition, invariant to image rotation, to affine illumination changes and robust to limited perspective deformations. Experimental results for retrieval show an excellent performance up to a scale factor of 4 for a database with more than 5000 images. Secondly, we automatically construct visual models for the retrieval of similar images. Models are constructed from a set of positive and negative sample images where no manual extraction of significant objects or features is required. Our model allows to efficiently capture "texture-like" structure and is based on two layers : "generic" descriptors and statistical spatial constraints. The selection of distinctive structure increases the performance of the model. Experimental results show a very good performance for retrieval as well as localization.