As a step toward computer assisted imagery interpretation, we are developing algorithms for computed radiography that allow a computer to recognize specific bones and joints, and to identify variations from normal in size, shape and density. In this paper we report on our approach to model-based computer recognition of hands in radiographs. First, image processing hypotheses of the imaged bones. Multiple hypotheses of the size and orientation of the imaged anatomy are matched against stored 3D models fof the relevant bones, obtained from statistically valid populations studies. Probabilities of the hypotheses are accrued using Bayesian inference techniques whose evaluation is guided by the structure of the hand model and the observed image-derived evidence such as anti-parallel edges, local contrast, etc. High probability matches between the hand model and the image data can cue additional image processing-based ssearch for bones, joints and soft-tissue to confirm hypotheses of the location of the imaged hand. At this point multipule disease detection techniques, automated bone age identification, etc. can be employed.