Protein molecules exhibit varying degrees of flexibility throughout their three‐dimensional structures. Protein structural flexibility is often characterized by fluctuations in the Cartesian coordinate space. On the other hand, the protein backbone can be mostly defined by two torsion angles ϕ and ψ only. We introduce a new flexibility descriptor, backbone torsion‐angle fluctuation derived from the variation of backbone torsion angles from different NMR models. The torsion‐angle fluctuations correlate with mean‐squared spatial fluctuations derived from the same collection of NMR models. We developed a neural‐network based real‐value predictor based on sequence information only. The predictor achieved ten‐fold cross‐validated correlation coefficients of 0.59 and 0.60, and mean absolute errors of 22.7° and 24.3° for the angle fluctuation of ϕ and ψ, respectively. This predictor is expected to be useful for function prediction and protein structure prediction when predicted torsion angles are used as restraints. Both sequence‐ and structure‐based prediction of torsion‐angle fluctuation will be available at http://sparks.informatics.iupui.edu within the SPINE‐X package. Proteins 2010. © 2010 Wiley‐Liss, Inc.