Because of its appealing simplicity, the anisotropic network model (ANM) has been widely accepted and applied to study many molecular motion problems: such as ribosome motions, the molecular mechanisms of GroEL-GroES function, allosteric changes in hemoglobin, motor-protein motions, and conformational changes in general. However, the validity of the ANM has not been closely examined. In this work, we use ANM to predict the anisotropic temperature factors of proteins obtained from X-ray and NMR data. The rich, directional anisotropic temperature factor data available for hundreds of proteins in the Protein Data Bank (PDB) are used as validation data to closely test the ANM model. The significance of this work is that it presents a timely, important evaluation of the model, shows the extent of its accuracy in reproducing experimental anisotropic temperature factors, and suggests ways to improve the model. An improved model will help us better understand the internal dynamics of proteins, which in turn can greatly expand the usefulness of the models, which has already been demonstrated in many applications.