To overcome the drawback of using only low-level features for the description of image content and to fill the gap between the perceptual property and semantic meaning, this chapter presents an object-based scheme and some object level techniques for image retrieval. According to a multi-layer description model, images are analyzed in different levels for progressive understanding, and this procedure helps to gain comprehensive representations of the objects in images. The main propulsion of the chapter includes a multi-layer description model that describes the image content with a hierarchical structure; an efficient region-based scheme for meaningful information extraction; a combined feature set to represent the image at a visual perception level; an iterative training-and-testing procedure for object region recognition; a decision function for reflecting common contents in object description and a combined feature and object matching process, as well as a self-adaptive relevance feedback that could work with or without memory. With the proposed techniques, a prototype retrieval system has been implemented. Real retrieval experiments have been conducted; results show that the object-based scheme is quite efficient and the performance of object level techniques have been confirmed.