The description and preliminary results of a computerized image analysis system which can analyze anatomical three-dimensional images and label each part of the image with the help of a knowledge base of human anatomy are presented. The approach utilizes a strong low-level image analysis system with the capability of analyzing the data in both bottom-up (or data-driven) and top-down (or model-driven) modes to improve the high-level recognition process. With the help of a multiresolution representation scheme, various cross sections of the input data set are first registered in proper orientation. Knowledge of reference organ parts is applied in 3-D volume representation to establish a correlation between the input data and anatomical knowledge base. The process of recognizing an object in the image is realized as hierarchical labeling of regions in the image. The control strategies are capable of postponing the recognition process of an object if the confidence in the matching process is not high enough to label the object according to the model.