The identification of DNA-binding proteins (DBPs) plays an important role in functional annotation of genes and proteins of prokaryote and eukaryote organisms. This study, for the first time, combined the gapped-dipeptide composition (GapDPC) and recursive feature elimination (RFE) to identify DBPs. The position specific scoring matrix (PSSM) of each tested amino acid sequence was obtained. Based on the PSSM, their GapDPC features of the amino acid sequences were extracted, and then the optimal features were selected using the RFE method. Subsequently, the support vector machine (SVM) was chosen as a classifier and the datasets PDB396 and LB1068 were tested using the jackknife cross validation test. The result showed that the values of accuracy, Matthews correlation coefficient, sensitivity, and specificity for the identification of DBPs were 93.43%, 0.86, 89.04% and 96%, and 86.33%, 0.73, 86.49% and 86.18% for the datasets PDB396 and LB1068, respectively, which were obviously superior to the methods reported previously in the literature. The new model established in this study improved the identification methods of DBPs.