We carry out an extensive statistical study of the applicability of normal modes to the prediction of mobile regions in proteins. In particular, we assess the degree to which the observed motions found in a comprehensive data set of 377 nonredundant motions can be modeled by a single normal-mode vibration. We describe each motion in our data set by vectors connecting corresponding atoms in two crystallographically known conformations. We then measure the geometric overlap of these motion vectors with the displacement vectors of the lowest-frequency mode, for one of the conformations. Our study suggests that the lowest mode contains useful information about the parts of a protein that move most (i.e., have the largest amplitudes) and about the direction of this movement. Based on our findings, we developed a Web tool for motion prediction (available from http://molmovdb.org/nma) and apply it here to four representative motions-from bacteriorhodopsin, calmodulin, insulin, and T7 RNA polymerase.In the analysis of protein dynamics, an important goal is the description of slow large-amplitude motions. These motions, while strongly damped, typically describe conformational changes which are essential for the functioning of proteins. Only global collective motions can significantly change the exposed surface of the protein and hence influence interactions with its environment. Such structural rearrangements in the protein can occur on a local level within a single domain or can involve large movements of protein domains in a multidomain protein. Protein dynamics thus cover a broad timescale: 10 −14 -10 sec (Wilcox et al. 1988). However, many large-amplitude conformational changes are not on a timescale accessible by most timedependent theoretical methods, such as phase space sampling techniques (e.g., molecular dynamics). Therefore, in order to gain insight into the mechanism of slow, largeamplitude motions, one must resort to the use of a timeindependent approach, such as normal mode analysis (Levitt et al. 1985).Normal mode analysis (NMA) is a fast and simple method to calculate vibrational modes and protein flexibility. In NMA, sometimes restrained to C␣ atoms only, the atoms are modeled as point masses connected by springs, which represent the interatomic force fields. One particular type of NMA is the elastic network model. In this model, the springs connecting each node to all other neighboring nodes are of equal strength, and only the atom pairs within a cutoff distance are considered.All existing NMA techniques have important common limitations resulting from the use of the harmonic approximation, the neglect of solvent damping, and the absence of information about energy barriers and multiple minima on the potential energy surface (Elber and Karplus 1987;Frauenfelder et al. 1988;Hong et al. 1990). In fact, the most interesting biologically significant low-frequency motions in a realistic environment are overdamped and hence not vibrational at all, rendering the corresponding normal mode frequencies of ...