Model-based automatic target recognition (ATR)systems deal with recognizing three dimensional objects from two dimensional images. In order to recognizeand identify objects the ATRsystem must have one or more stored models. Multiple two dimensional views of each three dimensional objectthat may appear in the universe it deals withare stored in the database. During recognition, two dimensional view of atarget object is used a query image and the search is carried out to identify the corresponding three dimensional object. Stages of a model-based ATR system include preprocessing, segmentation, feature extraction, and searching thedatabase. One of the most important problems in a model-based ATR system is to access themost likely candidate model rapidly from a large database. In this paper we propose new architecture for a model-based ATR systemthat is based on association-based image retrieval. We try to mimic human memory. The human brain retrieves images by association. We use generalized bi-directional associative memories to retrieve associated images from the database. We use the ATR system to identify military vehicles from their two dimensional views.