The B-factor, also called the Debye-Waller factor or the temperature factor, is a descriptor of the flexibility of protein and is commonly used in PDB (Protein Data Bank) format files. A B-factor could be measured from a protein crystal by x-ray scattering, but could not be got from the protein sequence directly. Thus, predicting the B-factor only based on the protein sequence could provide some references for the related researchers. In this study, we attempt to predict the B-factor based on the protein sequence. The information in AAindex and the predicted protein secondary structure, relative accessibility, disorder and energy changes are used to describe the amino acid residues. Four machine learning methods are used for modeling and prediction. The 5-fold cross validation is used to evaluate the modeling performance. As a result, this work provided some new methods for predicting and analyzing the B-factor based on the protein sequence, and we hope that this work could be helpful for the related researches.