Abstract. Navigation in computer aided orthopedic surgery often requires a 3D model that is calibrated to the anatomy of interest. We present a method for fitting a 3D model of the femur head region to 2D fluoroscopic X-ray images of the patient anatomy. This method doesn't require acquisition of a pre-operative CT. We create an Active Appearance Model from a training set of CT images of the femur head region. We then perform a procedure for searching for the AAM instance and the bone pose that best fit the patient's specific bone shape and pose. We measure similarity of the AAM instance to the patient's anatomy by comparing DRRs of the AAM to the fluoroscopic images. We use a gradient-descent algorithm to iteratively refine the AAM shape parameters and the bone pose parameters until convergence to a minimum of the dissimilarity metric.